Changelog

A record of what has changed in each Claira release, including new features, improvements, and fixes.

Release 061 -- October 22, 2026

A step-by-step Relevance Review guide that ends in a recorded result; bulk scan refunds that return to the tokens that paid, with a clear warning about scans that run past your renewal; field write errors that say why and what to do; field names Claira proposes that start with "🟢 Claira - "; and two new rows in the billing ledger.

Relevance Review workflow article

New workflow guide: Relevance Review. A step-by-step guide to running a first-pass relevance review with Claira: scoping the review set, running the early case assessment, building the Case Context, writing the prompt and testing it on a sample, getting sign-off from the lawyer or client, running the bulk scan, checking the uncertain and disputed documents, rerunning what was overturned, and reporting. See Relevance Review.

Help article corrections: skipped documents, error codes, Scan as default

Help articles corrected on skipped documents and error codes. Media scans now says that in a scan in your browser, a file the AI model cannot read (a PDF with no pages, a damaged or password-protected file) is skipped even in Auto and even when the document has extracted text. Bulk Scan lists the same cases under Skipped in its counters and document states. Error codes now says which E-FETCH and E-AI notes are worth retrying and which give the same result every time, that E-TOOLARGE marks a document skipped for oversized text and failed for a file sent with a very long prompt, and that a file format the selected model does not read (MODEL_MEDIA_FORMAT_UNSUPPORTED) is skipped for a PDF and failed for a multi-page scan.
Scan as starts on Auto in Single Review. The Single Review article said the default was Text; it is Auto, as Media scans already said.
The open document on a Claira Private case. The Agent Mode article now says that on a Claira Private case only the extracted text of the document you have open goes with your messages.

Field write errors say what to do

A failed write to a Nuix Discover field now says why, and what to do. When Claira cannot save an answer to a field, the message used to suggest that the field ID might be wrong, whatever had happened. It now names the cause: your Nuix Discover session has ended, your account may not code the field, the field's type does not accept the value (a pick list, date or number field), the field already holds a different value, or Nuix Discover could not be reached. Each message ends with the step to take. This applies to saving a single answer, to Multi-Code, and to the [E-WRITE] line of a bulk scan. See Error codes.

Refunds return to the tokens that paid, and a clearer monthly tokens email

A bulk scan refund now returns to the part of your balance that paid for it. Plan tokens come back as plan tokens and reset at your next renewal, like the rest of your allowance. Tokens you bought, and tokens billed as overage, come back as tokens that do not expire. Until now every refund came back as tokens that do not expire. If your plan renews before a task finishes, plan tokens from the month that ended are not returned. Tokens already on your balance are not affected. See Plans & billing.
The monthly tokens email shows the tokens that will not expire until used as their own figure. The email lists your plan's tokens for the month, the tokens that will not expire until used, and the total, so the three add up. When a case holds none, that line is left out.

Field names Claira proposes start with "🟢 Claira - "

Every field name Claira proposes now starts with "🟢 Claira - ". Claira's fields sit together in the Nuix Discover field list and are easy to recognize as AI fields. The Agent's field card proposes names such as 🟢 Claira - Privilege notes, the one-click logging setup creates 🟢 Claira - Log and 🟢 Claira - Prompt Log, and the Create in Discover form opens with 🟢 Claira - already typed. You can still type any name you like. Fields you already have keep their names, and the one-click setup still recognizes Claira Log and Claira Prompt Log rather than creating duplicates. See Creating fields with Claira and Fields.

The billing ledger shows overage refunds and tokens credited by Claira

Two new rows in the billing ledger. When Claira removes overage from a billing cycle, the ledger shows it as Overage refunded, with a minus, and the overage billed for that cycle falls by the same number. When Claira adds tokens to a case, the ledger shows Tokens credited by Claira. Your usage figures do not change in either case. Both can be picked in the ledger's type filter. See Usage.

Insight template articles follow the Quickstart dialog

The Insight template articles now describe the Quickstart dialog. Chronology, Key documents, Issue analysis and Privilege log told you to pick a template from a row at the top of the New insight card, which no longer exists. Templates sit behind Quickstart on the Insights page: pick a tile to see the prompt it reads and the prompt it will put in the box, then Use this template fills the prompt and selects a completed bulk task that ran the matching prompt, or Run this prompt opens Bulk Tasks with the prompt filled in. When no completed task has run the prompt, Claira warns you once a bulk task is the source rather than blocking the Insight. See Insight templates.

Relevance Review: record the result; plans and billing: refunds at renewal

The Relevance Review guide now says which figures to keep. A new Record the result part of Step 7 lists them: agreement on the sign-off sample, how many relevant documents turned up in the sample of the not-relevant pile, agreement on the relevant sample, and how each sample was chosen. It also says how to size samples with the lawyer, in proportion to the matter. See Relevance Review.
What a bulk scan refunds, under its own heading. Plans and Billing now says plainly that a bulk scan which ends after your plan renews does not get back the plan tokens it reserved and did not use, including the unused part of an Auto scan's hold, and how to check your reset date before a large scan. Bulk Scan and Media scans link to it.

Release 060 -- October 9, 2026

Very long prompts that are kept and open again, with a clear message when a prompt is too long for the Prompt Log; a clear message when a document's file is too large to scan in your browser; files the AI model cannot read skipped instead of failing a scan; and step-by-step help for giving a teammate access to the Admin Portal.

The open document's chip says Text on a Claira Private case

In Agent Mode on a Claira Private case, the chip for the open document now says Text. It said Text and file, but your own model reads text, so the document's own file was never used. Claira no longer downloads that file and sends it with every message, which makes a message quicker to send when the document is large. Hover over the chip to see why only the text goes. See Attach documents to a chat.

A document's file that is too large for a scan in your browser

A clear message when a document's file is too large to scan in your browser. A single scan, a Multi-Code run or a bulk scan in your browser reads files up to 23 MB, and a larger file used to fail with a generic error. Claira now tells you the file's size and the limit before anything is sent, and what to do next: run it as a background scan, which reads files up to 50 MB, or scan the document's extracted text. See File size limits and Scan says the document's file is too large.
Auto reads the text instead. With Scan as on Auto, a document whose file is over the limit is scanned from its extracted text at the Text rate instead of failing. In a bulk scan in your browser, an Image, Audio or Video run skips such a document with an E-FETCH note that gives the file's size, and does not charge it.

Giving a teammate access to the Admin Portal

The help site now has step-by-step instructions for giving a teammate access to the Admin Portal. They cover making someone an Organization admin on the Users screen, what that role lets them do, and how the rest of the team can still follow a case's tokens without the portal. See Give a teammate access to the Admin Portal.

A file the AI model cannot read no longer fails the scan

A document whose file the AI model cannot read is now skipped with a clear note. In a bulk scan that runs in your browser, a PDF with no pages, or a damaged or password-protected file, used to be tried several times and then marked failed with a generic error. Claira now skips it straight away, does not charge it, and writes an E-FETCH note that says the model could not read the file and how to scan the document's extracted text instead. The task's details count these under File the AI model could not read. See Bulk scan.
A clearer message when the AI model does not accept a document. When the model refuses a request it will refuse again, the note now says so, and Claira no longer spends time retrying it. See E-AI.

Very long prompts are kept and open again

A very long prompt is now kept, and opens again. A prompt longer than 100,000 characters, for example one with a whole report pasted into it, could be scanned but was not saved in the prompt box or added to Prompt History, so you could not open it afterwards. Prompts of up to 1,000,000 characters are now saved, listed in Prompt History and opened in prompt details like any other. A prompt of that length that you scanned before this release appears in Prompt History the next time you scan with it. See Very long prompts.

A prompt too long for the Prompt Log

A bulk scan whose prompt is too long for the Prompt Log now tells you, and runs without it. With Write the prompt to each document's Prompt Log ticked, a bulk scan with a very long prompt (about 300,000 characters or more) used to fail to start with a message that did not say why. The scan confirmation now says that the prompt is too long to write to each document's Prompt Log, shows the checkbox as unavailable, and the scan runs without it. The prompt is still kept in Prompt History and in the task's details. A scan the Agent starts for you shows the same message as it starts. See Very long prompts and Bulk scan.
The same limit applies when you save a single result. With Also write the prompt to the Prompt Log ticked, Apply in Prompt Lab, Save to fields in Multi-Code and saving an Agent answer to a field used to add the prompt to the document's Prompt Log whatever its length. A prompt over the limit is no longer added: the value is saved as before, and the message after the save says that nothing was added to the Prompt Log. See Recording the prompt in the Prompt Log.

Release 059 -- October 7, 2026

A deleted Agent Mode chat that is permanently erased after 7 days; pricing wording brought up to date in Claira, the Admin Portal, the help centre and the monthly token email; an Agent Mode that knows about the early case assessment as soon as Claira opens; and fixes for Claira Private and Multi-Code.

Agent Mode knows about the early case assessment as soon as Claira opens

When Claira opened, Agent Mode could fail to pick up whether the case had an early case assessment. In a case with none, the opening suggestion Give me a first look at what's in this case was missing. In a case with one, a bulk scan proposal for All documents did not show the assessment's document count and date, and pointed you to run an assessment instead. Both came right only after you opened Early case assessment from More. They now show from the moment Claira loads. The Agent's answers about the assessment were not affected. See Asking the Agent.

The monthly token email describes the reset correctly

The email that announces your plan's monthly tokens said the wrong thing about last month's tokens. It told you the new allowance was added to what the case already held, and that unused tokens roll over and never expire. Neither is how plans work: on your renewal date your plan's tokens are reset to the full monthly amount, tokens left unused in the month just ended do not carry forward, and only tokens you bought never expire. The email now says exactly that, in English, French and German. Nothing about your balance or your plan changed; only the wording did. See Plans & billing.
Monthly token emails appear in the Admin Portal delivery log. Filtering the log by Token updates now lists these emails; before, the filter came back empty for them. See Email routing.

Pricing wording brought up to date across Claira, the Admin Portal and the help centre

Leftover wording from the old token model is gone. A few places still spoke of a token reservoir, a daily refill, or tokens that roll over. They now describe plans as they work today: a monthly allowance that resets on the case's renewal date, with only bought tokens never expiring. Nothing about your plan, balance or bill changed.
In Claira: the downgrade confirmation now says your plan's tokens reset to the new plan's amount from your next renewal and that bought tokens are not affected; the message shown when you have already claimed today's free tokens speaks about you rather than the case, since each person has their own daily claim; and the Enterprise plan card no longer mentions priority capacity.
In the Admin Portal: the email family is named Token updates, as it is in Claira's own email preferences, with a description of when it is actually sent (once a month for each case, on that case's renewal date); the token balance tooltip on Cases, the overage permission and the overage rate help no longer mention a reservoir. See Email routing.
In the help centre: token packs added to a monthly invoice are described at the flat $0.10 a token every plan pays (Monthly invoice billing); the upgrade email is no longer said to name newly unlocked features, because every feature is on every plan (Plans & billing); Media scans and What Claira can do make clear that per-document token costs apply to bulk scans and single scans are free; Model selection no longer suggests the model changes what a scan costs; and Bulk scan explains that a short balance blocks a task only while overages are off.

Usage and billing help, and the Buy tokens dialog, describe the monthly reset

The Usage Analytics and Billing Analytics articles still said that unused tokens roll over and never expire, and that token packs are priced at your plan's rate. They now match how plans work: your plan's tokens reset on the case's renewal date and unused ones do not carry forward, tokens you buy never expire, and a token costs $0.10 in a pack on every plan. The Usage article also describes the reset date and purchased tokens notes shown under the balance. Nothing about your balance or your pricing has changed. See Your plan's tokens reset each month.
The Buy tokens dialog no longer says packs are priced at your plan's rate or that buying ahead costs less than overage. A token costs the same in a pack on every plan, the same rate as overage.
The same two articles also catch up on two details. The Claim button is described as yours: each user on the case can claim 100 free tokens once a day, up to the number of claims the case's plan allows per day. A scheduled downgrade is described as it works: from the next renewal, the plan's tokens reset each month to the new plan's amount, and tokens you bought are not affected.

A deleted Agent chat is permanently erased after 7 days

Deleting an Agent Mode chat now erases what it held. The chat still disappears for everyone straight away. Seven days later Claira permanently erases its messages, its question and search term cards, and the files and documents attached to it. Before, that content stayed in Claira until the whole case's data was deleted. See Sharing, pins, and branching.
Removing a file from a chat clears its text at once, along with the stored file. The file's name stays on the questions it was sent with, struck through.
Deleting a case's data removes attached files immediately. Files attached to chats that could not be read as text (a scanned PDF, an image) are now deleted from storage when the case's data is deleted, instead of expiring later.

Converting a prompt to Multi-Code works in a Claira Private case

In a Claira Private case, converting a Prompt Lab prompt to Multi-Code failed. The conversion asked for one of Claira's models instead of your own, and a Claira Private case runs only on your own model, so it was refused. The conversion now runs on your own model, like everything else in the case.

Claira Private no longer talks about tokens when a scan fails

A Claira Private case has no tokens, and three places still mentioned them. When a bulk scan could not finish a document, the note written to the field said tokens would be reimbursed; the progress panel's explanation of failed documents said the same; and each task showed a Tokens used count that was always zero. In a Claira Private case the note now says the model did not complete the scan and to run the failed documents again, and the token lines are gone.
The note written to the field when every retry is used up now appears in French and German too. It was English in every language.

A Multi-Code answer is shown under the name of the field it is written to

A new Multi-Code row starts on your working field. If you then pointed the row at another field by creating or connecting one from that row, the row kept the first field's name. The answer went to the right field, but the response and the live bulk results showed it under the old name. The row now takes the name of the field it writes to.

Release 058 -- October 4, 2026

A Case Context that keeps every fact of its documents and carries it in full into your prompts, with a summary, source numbers and a record of which version of the context each generated prompt used; one-click reuse of your Discover API token across cases, and an organization token in the Admin Portal; comments on bulk tasks, bulk scans linked to the prompt version they ran, and task results in one line; every past invoice as a PDF in the Admin Portal; very large bulk scans that start in seconds; and the retirement of Gemini Flash 2.5 and Gemini Pro 2.5, with every choice moved to Gemini Flash 3.5.

Partner discount set by Claira

The partner discount on the Admin Portal's Billing screen is now read-only for every admin. Claira sets it as part of your commercial terms; contact Claira to change it. You can still change the invoice window. See Billing and invoices.

Invoices dated the last day of their billing period

Invoices Claira issues to your organization are now dated the last day of the invoice window they bill, and payment is due 30 days after that date. Claira still issues and emails each invoice the morning after its window closes, once the usage is final, so an invoice that reaches you on September 22 is dated September 21. The PDF and the email say so in one line under the billing period, with the day the invoice was issued, and the Admin Portal's Invoice card now shows the Invoice date and the day it was Issued on. Invoices already issued keep their date. See The invoice date and the issue date.

Reuse documents across Case Contexts

A document can now source as many contexts in a case as you need. You can build a second version of a context from the same documents as the first, where Claira used to refuse with "already linked to another context". Deleting a context now only removes its own links to its documents; any other context they source keeps them. See Setting up Case Context.

Bulk scans linked to their prompt version

Every bulk scan is now linked to the version of the prompt it ran. In Prompt History, a prompt's details list Bulk scans with this version: every bulk scan in the case that ran that exact wording, whoever started it, including scans run before this release. Select one to open its task details. See Bulk scans that ran a version.
A bulk scan's task details now show Prompt version ("Version 4 of 5"), with the Final badge when the scan ran the version your team marked final, an amber note when it did not, and Open in Prompt History to jump to that version. See Which prompt version a task ran.

Bulk task results in one line

Each finished task's card in Bulk task history now shows a single result line, for example 1,372 succeeded · 66 failed, instead of a list of documents and error messages. Open the task to see why documents failed or were skipped: Why documents failed and Why documents were skipped count the documents by reason, such as Too large to scan or No text to scan. See Bulk Scan.

Comments on bulk tasks

Bulk tasks now have comments, like prompts do. Open a task from the Bulk Tasks history and use Comments, under its prompt, to note what the run was for: which sample it was, where its documents came from, what you were testing. Everyone working in the case can read the comments and add their own, and you can delete yours. A task with comments shows a speech-bubble icon and the number of comments on its card, and the Bulk Task History Report prints each task's comments under its prompt. See Comments on a task.
The Agent reads and writes comments. In Agent Mode, ask Claira what the team noted on a task or a prompt, or ask it to add a comment for you. Its comments carry your name and are marked Added by the Agent, on screen and in the PDF and spreadsheet exports, and you can delete them like your own. See Agent Mode.
Every bulk scan the Agent starts explains itself. The new task gets a comment saying why the Agent started it, what it is for and what the task does. The scan card shows that note under Comment on the task before the scan starts.
Deleted comments are recorded. Deleting a prompt or bulk task comment is now kept for your organization's compliance records (who deleted which comment, whose it was and when, never its text), and organization admins see it in the Admin Portal's Activity feed under Everything. See Deleted comments.

Every past invoice in the Admin Portal

The Invoices list on the Admin Portal's Billing screen now covers every invoice window since your organization's first case was activated, twelve at a time, with Show earlier invoices for older ones. Every window that has an invoice downloads as a PDF at any time, including invoices Claira sent you before it issued invoices in the portal, once Claira adds them. See Your invoice history.
A past window that closed with cases to bill but has no invoice now offers Request invoice. Claira is notified, issues the invoice in the portal dated the last day of the window, and emails it as usual. See Asking for a missing invoice.
Changing your invoice day no longer affects invoices already issued: they keep their own dates and stay in the list, and the first window after them starts where your last invoice ended, so no day is billed twice. See Payment methods and settings.

Early case assessment PDF looks like the screen

The early case assessment PDF now looks like the assessment in Claira: the four figures, The assessment in its box, and on its own page Data analysis with the same bars, timeline chart, colored tables and custodian coverage bars you see on screen, in the same order. Lists show the same first eight values and add up the rest under Other values, long names wrap instead of being cut, wide tables continue underneath, and the sample of titles comes last in two columns. See Downloading the PDF.

Reuse your Discover API token across cases

Background processing can now connect a Discover API token with one click. When your organization uses the Admin Portal, a token you paste in one case is offered back to you, and only to you, in your other cases of the organization (Use your API token); uncheck Also offer this token in my other cases to keep a token to one case. The bulk scan confirmation, Send to background and Field Migration offer the same one-click connection when a case has no token yet. See Connect with one click.
Replacing a token you use in several cases now updates all of them at once, or only the current case if you prefer. Disconnect from this case replaces Revoke and keeps your token saved for your other cases, and Remove my token from every case removes it everywhere. Disconnecting or removing a token, in a case or in the Admin Portal, now pauses the background tasks running in the affected cases at once, with their progress kept, and emails the person who started each one; connect another token, then use Resume. See Replacing, disconnecting or removing a token.
When Discover stops accepting a token, the background tasks that use it now pause instead of failing document by document: they show as Incomplete in Bulk Tasks, anything they already wrote to Discover is kept, and the person who started each one is emailed. The Background processing section shows Needs attention. Replace the token once, then use Resume on the paused tasks; Resume is refused until the case has a working token again. A field migration running at that moment ends as failed instead. See When Discover stops accepting a token.
Organization admins can set one organization token per Discover portal in the Admin Portal (Settings > Discover API token), which every case of the organization offers with Use organization token. The card shows how many active cases the token can open, the Personal API tokens option decides whether people may reuse their own tokens, and Switch to the organization token moves cases off personal tokens. The Cases screen shows which token each case uses. If Discover rejects the organization token, members whose role is Organization admin or includes Manage the organization's Discover API token are emailed once. See The organization's Discover API token.
When your organization enforces Claira roles, pasting a new token into a case is now checked too. The permissions are renamed Paste a new API token into a case and Disconnect a case's API token, and the new Connect a saved API token to a case, which every User has by default, covers connecting a saved one. See Users and Roles.

A detailed Case Context, carried in full into prompts

A Case Context now keeps every material fact of the documents you build it from: Claira removes repetition but no longer summarizes, so a long, detailed document gives a long, detailed context. After each update, Claira checks that every fact the context already held is still in it and puts back, word for word, any line the update left out. See From documents to prompts.
Every context now opens with a Summary: up to three sentences on why the review exists. Claira writes it from the sections, so you cannot edit or remove it; change the sections and Claira rewrites it. A context created before this release gets one the first time you open it, and the summary opens the downloaded PDF and Word files. See The summary.
Each line Claira writes from a document now ends with that document's number in the source documents list, such as [2]. Click a number to see its document in the list. Lines your team or the Agent added carry no number, and lines written before this release have none. The downloads show the same numbers. See Source numbers.
The Prompt Generator now carries every fact your task uses from the Case Context into the prompt in full, so a long context gives a long prompt, and shows three steps while it works: Reading the Case Context, Writing the prompt, and Checking every matter fact is in, which adds any fact your task uses that the draft left out. Use with context on a Quickstart template shows the same steps. See Describe a prompt with the Prompt Generator.
In Prompt History, a prompt the Prompt Generator built from a Case Context now says which context and which version of it, and a note tells you when that context has changed since. See Prompts written by AI.

Gemini 2.5 models retired

Gemini Flash 2.5 and Gemini Pro 2.5 have reached the end of their life and are no longer in the model picker. If you had chosen one, your choice moved to Gemini Flash 3.5 automatically, with your prompt, fields and model settings unchanged. Bulk tasks already running when the update arrived finish on the model they started on; every task you start or resume afterwards runs on Gemini Flash 3.5. Agent Mode, Case Context, Insights and early case assessment now run on Gemini Flash 3.5 in every region. Finished scans and tasks still show the model that answered them. For harder documents, raise the Reasoning level in Model settings. See Retired models.

Large bulk scans start quickly

Starting a very large bulk scan no longer keeps Claira busy for minutes. A scan of 150,000 documents now gets its Task ID in seconds, and Scan List keeps responding while another scan in the same case is starting.
If starting a scan takes longer than usual, for example on a slow connection, Claira now checks whether the task started instead of reporting a failure. When it finds the task, the scan carries on as normal. If it cannot confirm it, it asks you to check your bulk tasks before you start the scan again, so the same documents are not scanned twice by mistake. See Monitoring background tasks.

Release 057 -- September 29, 2026

New pricing that gives every plan far more tokens, a monthly allowance with bought tokens that never expire, 100 free tokens a day for each person and free Insights, a Team plan and Claira Private, and monthly invoices that Claira issues and emails itself; Insights that read your fields as they are when you generate them, appear as they are written, and come as tables, timelines, Word and Excel files; an Agent Mode that reads the open document, takes attachments and can act on its own with Give permissions; background processing for image, audio, video and Auto scans; early case assessment, organization prompt templates and search term families; a Case Context that reads scanned documents and recordings and saves as you type; and a redesigned Claira that fits a pane as narrow as 320 pixels, alongside a long run of fixes to bulk tasks, Multi-Code and Field Migration.

Help desk language switcher

Switching the help desk to another language now keeps you on the article you were reading, instead of sending you to a page that does not exist. Previously, picking a language from the sidebar menu added a language code to the address instead of replacing it, so switching to German and back to English landed on a "page could not be found" screen.
Old links of that shape still work: they now redirect to the right article in the language you asked for, or to that language's help desk home when the article has not been translated.
The page now also reports the language you are reading to your browser and screen reader after you switch, and searches in that language.

Case Context reliability

Case Context: generating a Case Context from selected documents could fail with an "invalid response" error after a long wait, most often on cases whose documents quote names or phrases. Generation now completes reliably, and no longer needs a retry to succeed.
Case Context: when the selected documents are too long to process together, Case Context now says so and asks you to remove a document, instead of showing a generic "Unable to generate" error.

Germany opens in German

Claira now opens in German for customers on the German address. Sign in for the first time and the interface is German straight away, whatever language your browser is set to. Previously Claira started in English unless the browser itself was set to German. See First-Time Orientation.
Your own choice still wins. Pick English, Français, or Deutsch from the language dropdown in Settings and Claira stays in it. The choice is saved to your account and follows you to your next browser or device. The regional default only decides what you see before you have chosen.
Notification emails follow the same default. A message Claira sends from the German address to someone who has not picked a language yet now arrives in German rather than English.
Canada and Australia are unchanged and still open in your browser's language. Your language and where your data lives remain separate choices: every region offers all three languages, and data residency still follows the Claira address you open.

When a model declines a document, Claira now says so

A model refusing a document is no longer reported as "something went wrong". Some models apply their own content policy to the request and reject it before writing a single word. Claira used to show the generic failure message, which invited you to run the scan again. The scan failed again, identically, every time. It now names what happened and marks the document E-SAFETY, the same code a safety-filter block already carried.
The fastest fix is one click away. Content filters differ from model to model, so a document one model refuses is often scanned normally by another. On a single-document scan the error message now carries a Change model shortcut to the picker in Settings.
Bulk scans stop wasting attempts on it. A refused document is marked and moved past instead of being retried, so a corpus with a few refusals no longer slows the whole task down.
A stray \n\n no longer appears in the AI Response box when a scan fails partway through.

Export your whole prompt history for the record

The complete prompt history of a case now exports as a spreadsheet. Prompt History has an Export history button beside the user filter: every prompt, every revision of it, and the comments left on them, as Excel (.xlsx) or CSV. Until now only a single prompt could leave, and only as a PDF.
It answers "what was tried, and why did we trust this one?" Each revision arrives on its own row with its author, when it was first and last used, how many times, whether it carries the final mark and who set it. The row also shows how many words changed since the previous revision and the text that changed. The reasoning behind a revision travels with it: comments get their own sheet, joined to the exact revision they were left on by Prompt ID.
Export everything, or just what you are looking at. The scope choice appears when a search or user filter is active. A search matches whole chains, so a prompt never arrives with revisions missing from the middle of it.
The file discloses its own gaps. If comments could not be loaded, the summary sheet says so rather than leaving you to notice. All dates are UTC in ISO 8601 so they sort correctly and read the same in every region.
Every export is recorded for compliance (who, what format, how much, and when), written before the file is produced, and holding no prompt or comment text.

Create a Case Context without typing a name

You no longer have to type a name for a new context. When you create a context, the Name field arrives filled with the template's name (for example Review Mandate, or Review Context when you start blank) and selected, so you can create straight away or type over it. A name you typed survives a change of template. If you clear the field, the context is named after its template, and a name already used in the case gets a number (Case Context 2), with a note beside the field saying so. You can still rename the context later.

Gemini Flash 3.5 comes to Australia

Australian tenants now start on Gemini Flash 3.5. The newest Gemini Flash generation is now served from within Australia, so it is the default scan model there, as it already was in Canada and Germany. It brings a much larger context window than Gemini Flash 2.5 (useful for long documents that previously had to be truncated), and it runs inside the Australian data residency boundary like every other model Claira serves you. See Model Selection.
Gemini Flash 2.5 is still there. Existing cases keep the model they were set to; only new cases start on 3.5. If you would rather stay on 2.5 for a review already in progress, it remains in the picker.
Gemini Pro 2.5 has left the Australian picker. That model is no longer served from within Australia, so it is no longer offered to Australian tenants and any case still set to it now shows Gemini Flash 3.5. Canadian tenants are unaffected and keep Gemini Pro 2.5.
Agent Mode, Case Context, Insights and the prompt generator moved up with it. The features that do not use the model picker now run on Gemini Flash 3.5 for Australian tenants.

Insight reports link their own citations, and hovering a Document ID shows what it said

Citations in the exported PDF are now live links. Click a citation in the Insight body and the PDF jumps to that document's entry in the Source documents section. Verified citations are underlined so you can tell at a glance which ones will move; a citation Claira couldn't match to a source document isn't underlined and doesn't click, so the absence of a link is itself a signal. Each source entry gains a Cited on p. N line whose page numbers link back to where the document was cited, so a chronology can be checked line by line without leaving the file. Documents that fed the Insight but were never cited stay listed, tagged Not cited in this revision. The links work in any PDF reader (no Claira, no Nuix Discover), which is what makes the export reviewable by someone who was only sent the file. See Insights.
The PDF now prints the search that finds these documents in Discover. A Finding these in Discover panel at the top of the Source documents section names the Log Field and the reference stamped on every source document (for example IN012), so the search travels with the file instead of living in someone's notes. It replaces the bare reference that used to trail the header's Source line. See Insights.
Hover a Document ID in the Insights pane to see what it said. Hovering a verified citation (or any document in Show sources) opens a preview card with the document's title and the AI response that put it in the Insight. That response is stored on every Insight but had never been visible in the app; reading it previously meant downloading the PDF. Citations that couldn't be verified have no preview, because there's no document behind them. See Insights.

The Comprehensive PII (surface forms) prompt no longer assumes North America. It used to list SSN, SIN, passport numbers, driver's license numbers and illustrate formats with Montreal phone numbers and addresses, which meant identifiers like an Australian Tax File Number or Medicare number, or a German Steuer-Identifikationsnummer, could be passed over. The prompt now describes government-issued identifiers by what they do (tax, social insurance, national identity, health insurance, passport, driving licence), so an identifier counts whatever country issued it and whatever it is called locally. See Comprehensive PII (surface forms).
Tell Claira which country your documents come from. The prompt now carries an EXPECTED IN THIS COLLECTION block where you name the jurisdiction, and the help article has ready-made blocks for Canada, Australia, the United Kingdom and Germany/EU that you paste straight in. Naming the identifiers your documents actually contain improves results on schemes that otherwise read as ordinary reference numbers. Delete the block if you would rather not narrow the prompt: it works either way.
The phone and address rules now state the underlying principle (the same number or address written two ways is two entries) instead of demonstrating it with three North American phone numbers, so the rule reads the same wherever your matter sits.
New template: Special category data (GDPR Article 9). Where the PII template inventories every written form of every identifier, this one answers a narrower question: does this document touch racial or ethnic origin, political opinions, religious or philosophical beliefs, trade union membership, genetic or biometric data, health, or sex life and sexual orientation (and which of them). It judges content rather than wording, so a sick note counts as health data whether or not it says "health", and it reports criminal convictions (Article 10) alongside. Returns the category labels plus one sentence naming the supporting passage, and pairs well with Multi-Code when you need a field per category.
German fix: the PII Identification workflow article showed a prompt ending in KEINE PBD ERKANNT where every other copy of it says NO PII DETECTED. Anyone who copied that version got a "nothing found" marker their exports and Search Term Families splitting would not recognize. Both German articles now carry the same prompt as every other language.

Chat organization: rename, search, archive, and ordered pins

Agent Mode chats can now be renamed. Use Rename in a chat's ⋯ menu, or click the title in the top bar while the chat is open. Once you name a chat yourself, Claira leaves it alone: no automatic title will replace it later. See Sharing, pins, and branching.
Search now reads what was said in your chats, not just their names. Clicking the search field at the top of the chat list opens the full Chats view, where a query matches titles and the messages inside every chat you can see, showing a short highlighted excerpt of the match. It only ever looks inside chats you can already open: a teammate's private chat never appears.
Finished conversations can be archived from the ⋯ menu: the chat leaves your list but keeps its transcript, sharing, and pins, and the Archived filter in the Chats view brings it back.
Pinned chats can be put in the order you want: drag them in the Pinned section, or use Move up / Move down in the ⋯ menu. Personal pins are ordered for you alone; a team pin's position is shared with the whole case.
Chats that are not pinned are now grouped by when you last used them (Today, Yesterday, Previous 7 days, Previous 30 days, then by month), so a case with months of history stays scannable.

Claira can see your Insights

Ask Claira about the insights in your case. In Agent Mode, Claira can now list what has been generated and read one back to you (by name, or by its IN reference). It sees the whole case team's insights, not just yours, and it says so when it has only read part of a long one. Citations it cannot match to the source corpus are called out rather than repeated as fact.
Set an insight up in conversation. Describe what you want and Claira gathers the pieces (a finished bulk scan or a field across the documents you have selected, plus the instruction), then offers a card with a Generate insight button. Nothing is generated until you press it, and editing anything asks you to confirm again. The same card revises an insight you already have. See Insights.
Renaming an insight through Claira is free, as it is in the Insights panel.

Move a running scan to the background

A live bulk scan can now be handed to Background Processing while it is running. A Send to background button sits next to Cancel; click it and Claira finishes the documents already under way, then passes the rest to the servers. That way, you can close your laptop without cancelling and starting over. See Moving a live scan to the background.
The run keeps its same Task ID, so the Log Field reference (for example SC067) still finds the whole scan in Nuix Discover, and there is no second charge: the tokens for the run were reserved when it started. Documents already coded are never scanned again, and documents that failed during the live part are retried on the servers.
The button explains itself when it can't be used: hover it and Claira names the reason (most often that Background Processing isn't set up for the case, or that the run is on a model background processing can't run). Image, Audio, Video and Auto scans can be sent too: Claira uploads the remaining documents' files first. See Running media scans in the background.

Usage figures no longer count the tokens cleared when a case is deactivated

Tokens used now counts only tokens spent on work. When a case was deactivated, the unused tokens cleared from its reservoir were counted as tokens used for that month: in the Usage mode total, the per-user chart, the Billing Report PDF and the monthly usage summary email. They are no longer counted, and past months have been corrected. See Usage Analytics.
The tokens-by-category chart now subtracts refunds, like the total. A partially refunded bulk task could make the category bar in Usage mode and in the Billing Report PDF longer than the month's tokens-used total. The categories now never add up to more than the tokens consumed, for past months too.
The monthly usage summary email now subtracts refunds from each case's tokens used, as Usage mode already did.

Applied revisions are kept in Agent Mode

Applying a Smart Revise suggestion no longer changes your prompt silently. Claira records it in an Agent Mode chat (what you asked for, what changed, and the revised prompt itself), and the dialog offers View this revision in Agent chat. One revise session gets one chat, so applying suggestions to three Multi-Code fields gives you a single thread with three entries. See Rate responses and Smart Revise.
The revised prompt goes back where it came from. Re-inserting from that chat returns a field's prompt to that field rather than to the AI Review prompt, and the insert now saves a prompt-history version like every other insert does. If the field has since been renamed or removed, the entry stays readable and only the insert button disappears.
Insight revisions are recorded the same way. Every revision in an insight's chain lands in one Agent Mode chat for that insight, whether it was run from the Insights panel or approved on a card in chat. A revision run from chat is badged Agent in the revisions sidebar, so you can always see where one came from. See Insights.
A chat that approves an insight card points to the result. Approve an Insight or one of its revisions on a card in any other chat, and that chat keeps a short note with an Open revision chat button that takes you to the Insight's own chat. See Insights.
Insight revision chats are easy to spot. Each is named after the Insight's reference (for example Insight revisions · IN012) and adds the title once there is one, and an Insight created from a card in chat is badged Agent from its first revision.

Agent Mode reads the document itself

Agent Mode now reads the open document, not just its extracted text. When you ask about the document you have open in Discover ("what does this document say?", "who signed this email?"), Agent Mode now reads the document itself (the PDF, image, audio or video) as well as its extracted text. You can now ask directly about scanned pages, handwriting, stamps, photos and recordings that have little or no extracted text. See Agent Mode.
It only does this when you refer to the document. General questions are answered as before, and files larger than 15 MB are read from their extracted text only.

Case Context PDFs show approval status and who approved

An approved Case Context now says so at the top of its PDF. The title carries a green APPROVED badge once the context is approved, and an amber DRAFT badge until then, so a printed or forwarded copy shows its status at a glance. A long context name now wraps before the badge instead of running underneath it. See Setting Up.
People are named in full. "Exported by" and "Approved by" now show each person's name next to their username instead of the username alone: for example, Jane Doe (jdoe). If a name cannot be looked up in Discover, the username still appears on its own. See Exports.

Running low on tokens, and the monthly usage email

A blocked scan now offers the fast ways out. When a bulk scan is short of tokens, the dialog offers Claim free tokens and Buy tokens alongside the plan manager, in order of what it costs you.
The monthly usage summary email now reports the tokens granted to each case (the plan's monthly tokens plus any free daily claims), where it used to show the daily refill.

Attach documents to an agent chat

You can now drop your own files into an Agent Mode chat. Drag a spreadsheet, PDF, Word file, text file or image onto the chat (or use the new paperclip beside the message box) and ask questions about it alongside everything Claira already knows about the case. Up to 5 files per chat, 10 MB each. See Attach documents to a chat.
The top bar now shows what a chat can read. A paperclip with a count opens a list in two groups: the files you attached (each removable), and (outlined in green) the Nuix Discover documents this conversation has already been shown. A Discover document appears there once Claira has actually received it (its text, its own file, or both), and hovering over it shows which, so the list reflects what the chat could actually use rather than whatever happened to be open.
Most files never leave your computer. Spreadsheets, Word files, text files and PDFs with a text layer are read in your browser, and only the extracted text is sent. Files with no readable text (a scanned PDF, an image) go to Claira's secure storage in your region so the model can look at them directly, and are deleted the moment you remove the attachment.
Attachments travel with a branch, so a branched chat answers the same way as the one it came from. Older .xls spreadsheets are not supported. Save as .xlsx or .csv first.

Claira remembers the field you are working in

Pick a field once and the other modes start there. Claira now remembers the destination field you last chose in a case and offers it as the starting point everywhere else: pick a field for a bulk scan and your next single review opens on it, and the same field seeds a new Multi-Code row, the From side of Field Migration, the source picker on a new Insight, and the destination on an Agent Mode bulk scan proposal card. Generating an Insight from a field sets the memory too. Before this, each part of Claira answered the question separately and you re-picked the same field several times a day. See Bulk Scan.
It is personal, and it is only a starting point. The remembered field is yours alone, per case: your colleagues each keep their own and your choice never changes what they see. Every picker stays fully editable, and changing it has no effect on a scan that has already run. Field Migration and Multi-Code only read it, so nothing you set up there changes where your next scan writes. See Field Migration.
A field picked with an empty prompt box is now saved. Choosing a destination field before writing your prompt used to be lost the moment you switched modes or reloaded, because the choice could only be saved together with a prompt. It is now saved on its own, so you can open a case, pick your field, and write the prompt afterwards.
Nothing is ever auto-picked for you. If the field Claira was holding has since been disconnected from the case, every picker opens empty and waits, including the one on an Agent Mode bulk scan proposal card, which asks you to choose a destination before Approve and Start will run. Claira does not quietly substitute another field: a bulk scan written into a field you did not choose is far more work to undo than one more click. See Agent Mode.
The field picker reads more easily. In Prompt Lab and Bulk Scanning, the "Used" and "Current" text labels beside a field are now small icons with tooltips: a circle with a check on a field you have used before in this case, an empty circle on one you have not, a clock while the field is waiting on access from your case admin, and a star on the field set in Settings. The labels used to take room from the field name in a narrow pane; the icons do not, so long field names stay readable.
"Scan as" now says why it is showing Text. When you have your own model selected, Claira can only send it text, so Scan as shows Text with a short note beside it giving the reason and telling you that your own choice comes back as soon as you switch to a Claira model. The control was already locked in that situation, but the explanation sat in a tooltip you had to go looking for. Your saved choice is never overwritten. In Local Mode, the note that was already beside the control says the same thing. The hint appears wherever you pick a scan type: Prompt Lab, Bulk Scanning and Multi-Code. See Media scans and Bring Your Own Model.

Every PDF Claira produces now numbers its pages. The prompt export, the full prompt history, Insight reports, the Bulk Task History Report, the Billing Report and the Case Context export all print "Page 3 of 12" at the foot of every page, opposite a line naming the report, the case and the date it was exported. A page that comes loose from a printed stack, or is scanned on its own, still says what it is and which matter it belongs to. See Exports.
Continuation pages repeat the report title. From the second page onward, a small line at the top of the page names the report, so someone who opens a long export halfway through knows what they are reading. The first page is unchanged: it already carries the full header.
A long title now wraps instead of running off the edge of the page. A long prompt, Insight title or Case Context name used to be drawn on a single line and vanished past the right margin. It now wraps onto up to three lines and ends in an ellipsis if there is more to say, and on the Case Context export it wraps clear of the APPROVED or DRAFT badge instead of running underneath it. See Insights and Prompt History.
Long task names in the Bulk Task History Report wrap onto two lines. A long name used to be cut short, which often left two task cards that read identically; names now wrap, with the media type and status pills still held to the right.

Bulk scanning says what you have selected, and past tasks are easier to find

The header reads your selection back to you. Where Bulk Scanning used to show a green "0 docs" ticker with nothing selected, it now says Select one or more documents to get started, and 3 documents selected once you have picked some, with correct plurals in English, French and German. On a narrow pane it shortens to the number alone; hover it to read the whole sentence. See Bulk Scan.
Bulk Scanning, Ongoing background tasks and Bulk task history now share one header: a green icon tile, the title, a help tooltip, and on the right whatever that card has to report (your selection, or live progress while a scan runs). The two lower cards had no help text at all before. Both have one now.
Finding a past task is a first-class thing. The history list has a search box that matches the task name, the person who ran it and the task number in either spelling (014 and 14 find the same task), and a funnel icon beside it that filters by who ran the task. The icon carries a count of how many filters are on, and Clear filters sits at the bottom of the panel. It replaces the Filter by user button in both places it appeared: the history card and the Task History dialog. See Finding a past task.
Filtering by a colleague whose name contains a space now works from the checkbox label. For exactly those names, clicking the label did nothing before.
The Background and Cancelling tags on a task explain themselves. Each now has a tooltip you can reach from the keyboard instead of a browser tooltip that only appeared on hover, and the status tag in the Task History dialog uses the same colour vocabulary as the rest of Claira. A task row joins its number and its name with a middle dot.
A running scan announces itself properly. A screen reader now hears "Bulk scan 40 percent complete" rather than a progress bar and a bare number read separately, and the count on Ongoing background tasks reads as "4 scans running".
The Bulk Task History Report opens on the most recent task and works backwards, instead of starting at the oldest run in the case, so the run you are most likely looking for is on page one.

Insights: a Quickstart for templates, the prompt beside the answer, and citations Claira stands behind

Insights is now one card, with the green mode header, a help tooltip explaining what an Insight actually is, and Past insights as its own list below the composer. See Insights.
Templates moved behind a Quickstart button. The five pills across the top are gone. Quickstart opens a dialog with one tile per template, each carrying a one-line description of what it produces and a Ready, Partly ready or Not run yet chip saying whether this case already holds the data it reads. Pick a tile to see the prompt it will use and which bulk run it matched; Use this template fills the prompt and selects that run, and the template then sits in the card as one chip you can remove.
A template whose prompt has never been run no longer blocks you. Claira warns instead, and offers to open Bulk Tasks with that prompt filled in, or to pick a field instead. Generating anyway is your decision.
The source row reads as a sentence. Under Use AI results from you choose A bulk task or Selected documents, and the two dropdowns line up instead of one being pushed down by a paragraph. A bulk task option now shows its SC reference and name, then the field it wrote to and how many documents the Insight would read, for example 113 of 115 documents. The number of selected documents is the same pill used everywhere else in Claira, and the prompt box is labelled Insight prompt.
The viewer shows the prompt that produced what you are reading: Original prompt on the first revision, Revision N request after it, three lines with the whole text on hover. On a narrow panel, where the revisions sidebar has to make way, a Revision dropdown above the text offers the same choice, so revision history is reachable there for the first time.
Citations that could not be verified are gone, and so is the idea of one. No more red struck-through citations in the viewer or the PDF, and no more "Citations verified: n of m" footer line. If the model names a document the Insight never read, Claira removes that citation before you see it and reports it to us as a fault to fix, rather than printing it for you to catch. Click a citation and Show sources opens and scrolls to that document's entry.
Show sources now matches the PDF: a numbered entry per document with its full title wrapping, its Document ID, a Not cited in this revision tag where it applies, the AI response beneath, and the search that finds those documents in Discover. A long list renders fifty entries at a time with a button for the rest.
Multi-Code results no longer arrive as raw data. Each field's name is shown in bold with its value, in the hover card, in Show sources and in the PDF, and the model is given the same labelled lines instead of the shorthand Claira stores the answers under.
The exported PDF carries the prompt in a panel under the header, and the header says Original or Revision 3 instead of a bare number. A long Insight title wraps in the viewer rather than being cut off.
Past insights gained a filter: by who created the Insight, by status and by date, with a count of applied filters and Clear filters. Dates read as 2 days ago while they are recent, and a row's revision count explains itself on hover.
Deleting an Insight asks first, in an in-app dialog that says what will be removed, stays open while it works, and confirms with a message when it is done. See Deleting an Insight.

Field Migration is now four guided steps

Four numbered steps: Fields, Options, Review and Try on one document, each opening when the one before it is answered, with Start migration in a bar pinned to the bottom of the screen. When Start is greyed out, the sentence beside it names the one thing still missing, and a collapsed step's heading says what you chose, so you can check your settings without opening anything. See Field Migration.
Only the From field is shown when you arrive. The To field appears once you have chosen what to migrate from, and a destination you already picked survives clearing the source. See Step 1: Fields.
The To list is grouped by how well the two field types fit: Best match, Compatible, Works with limits, Not verified, each with its own icon. Nothing is hidden, and compatibility is now one line under the picker with a Why? you can hover, instead of a coloured box.
Copy or move and "If the destination already has a value" are dropdowns whose chosen option keeps its consequence written underneath, so the caveat about Move clearing the source stays in front of you.
Splitting a memo value on a separator shows a live preview of what that separator does to a real value from your selection, and is not offered at all when the destination can hold only one value: a single line explains why instead.
There is no Check selection button any more. Claira checks your selection by itself a moment after any change. When you change a setting, the numbers you were reading stay on screen, greyed and marked Out of date, rather than vanishing, and Check again refreshes them or retries a check that could not finish.
The one-document test keeps its Undo. After a test, a banner names the document by its Document ID and keeps Undo test and Keep change on screen however much else you change; before this, any unrelated change took the Undo away. A second test waits until you have chosen one of the two, and starting a migration with a kept test change warns you in the confirmation and then carries on.
A run in flight locks the four steps read-only and says which migration is running. When it finishes, the button becomes Start another migration: your fields and options are kept, and a fresh check is required, so a second run can never start from the first one's numbers.
Migration history gained a filter for status, date, field (matching migrations where that field was on either side) and who ran it, with a count of applied filters and Clear filters. "No migrations match your filters" is now a different message from a case with no migrations at all, and a history that fails to load offers Try again instead of an empty list.
The selection count in the Field Migration header is the same pill as everywhere else, with correct plurals and without the old "in Discover" wording.

Prompt Lab: a quieter history list, prompt details worth opening, and a PDF that leads with the final wording

The history list is one card per prompt, and it is much quieter. A card leads with the prompt itself, in full ink over two lines, instead of a truncated hundred-character extract, and under it sits a single soft line of detail: who wrote it, how many versions it has, how often it has been used, and how many comments it carries. Pinned and final prompts show a small icon at the front. Dates have left the list and moved into prompt details, where you go when you actually want them. The grouped view and the chronological view now draw the same card, so a prompt reads the same way whichever one you are in. See Prompt History.
Click a card to open it. The View Prompt button on every row is gone: the card itself opens prompt details, from the mouse or from the keyboard. The chevron at the end of a card now does one thing only, showing and hiding that prompt's versions, and clicking a version opens that version's details.
Grouped or chronological is a quiet dropdown reading Grouped or Chronological, in place of the pair of bright green buttons. It still remembers the view you last chose.
One funnel icon holds all the filtering, with three sections inside it: the people who wrote the prompts, a date (Any time, Last 7 days, Last 30 days, Last 90 days) and a status (Pinned, Marked final). The icon carries the number of filters you have applied, Clear filters sits at the bottom of the panel and deliberately leaves your search box alone, and a search and filter combination that matches nothing now says so instead of showing you an empty list. The old Filter by user button is gone.
Export history is an icon in the toolbar now, with a tooltip saying what it does, rather than a labelled button competing with your prompts for attention.
Download PDF has moved into prompt details. It is no longer a button on every row of the list: open the prompt you want and it is in the header, still offering This version only or Full version history. See Downloading a prompt as PDF.
Downloading one prompt no longer looks like downloading all of them. The waiting state used to appear on every download control at once; it now belongs to the download you actually started.
The version you marked final sits at the top of the version list, with the rest below it newest first, so the wording the case settled on is the first thing you see. Numbering still counts from the first wording written, so Version 2 always means the second thing written wherever it now appears, and the arrows in prompt details still walk the versions in the order they were written. See Marking a final version.
The mark-as-final button explains itself. Hovering it now says what marking does (it records the wording the case settled on, everyone on the case sees it, and only one version can carry it) and, on the version that carries the mark, what removing it would leave behind. The Final badge is explained too.
Prompt details is now the place to look. Its title reads Version 3 of 5, with arrows to step to the previous or the next version without going back to the list, and under the prompt text there is one quiet line with the authors, when it was first and last used, and who marked it final. The Case row is gone, since it was always the case you are already in. See Prompt details.
The full version history PDF opens with the wording that matters. It now leads with Final version (or Current version, when nothing is marked) printed in full with its own comments, then Version history, newest first with each version's comments underneath it. The featured version keeps its place in the sequence and points back to the top instead of being printed twice, and the single pile of comments at the end of the document is gone.

An Insight now says how many documents it read

The line under an Insight's title used to read 3 documents selected, which was wrong twice over: an Insight built from a single document said "1 documents selected", and what it counted was never your current selection but the documents that Insight actually read when it was generated. It now reads 1 document read or 7 documents read, with correct plurals in English, French and German, both in the viewer and on the exported PDF. See Insights.

Usage calls the first bulk type "Bulk single", names your plan, and fits a narrow panel

The first bulk type is now called Bulk single. The Usage chart and the Billing Report PDF used to call it "Bulk one-field". Both now say Bulk single, beside the unchanged Bulk multi-code, and a ? beside the chart's title explains what each type counts. See Tokens by bulk type and media.
Usage now has the same header as the other modes: the green icon tile, the title Usage statistics, a help tooltip, and Refresh as an icon with a tooltip, unavailable while the figures are loading. See Usage Analytics.
The Plan tier card shows the plan's name. It used to print the plan's internal code in lower case. It now names the plan, and when the case upgraded partway through the cycle it names the highest plan held, which is what the Plan column of the Billing Report already showed. See Monthly summary cards.
Usage follows the width of the Claira panel, not of your browser window. The four monthly cards sit in one column on a narrow panel, then two by two, then in one row, and the token balance, the claim and buy buttons and the report buttons stack together when space runs out. Under the stacked chart, a long bulk type name wraps onto a second line instead of disappearing.
Names in the per-user chart no longer crowd out the bars. The column of names used to take a fixed width, as much as two thirds of a narrow panel. It now takes only the room the names need, a name too long for it ends in an ellipsis, and hovering its bar shows the full name and Tokens used. See Tokens consumed per user.
The month arrows have names. Previous month and Next month now carry tooltips and are announced by screen readers, which also hear the new month when you move.
Only the report you asked for shows that it is working. Downloading one Billing Report used to show "Generating report..." on both buttons. Now the button you pressed reads Preparing the download and the other waits until it finishes. Before the month has loaded, the month button reads Download billing report, where it used to read "Download Download billing report". See Billing Report PDF.
Dates and numbers follow the language Claira is set to, on the Usage screen and in the Billing Report, instead of your browser's settings: month names, renewal and claim dates, token counts and the pie chart's percentages. In French, for example, a share now reads "12 %".

Agent Mode cards say where they stand, and the chat list filters like the rest of Claira

Every proposal card states its status in its header. A bulk scan card reads Queued once you approve it; an Insight card reads Needs details, Generating or Generated; a card that creates a field reads Created or Awaiting admin access; and any card reads Failed or Ended once its work is over. Warnings and errors on the cards, such as no destination field being available, a field name already in use, or a selection that changed after the card was made, now appear as coloured notices. See Agent Mode.
Cards show your selection the way the modes do. The bulk scan card, and an Insight card over the documents you have selected, now carry the same pill as the rest of Claira in their header (3 documents selected, or Select one or more documents to get started), with correct plurals and the live count, in place of a sentence that did not handle a single document. An Insight card over your selection names the field it will read under Read AI results from.
The bulk scan card's destination field uses the same picker as Prompt Lab and Bulk Scanning, so a long field name is shortened and shows in full on hover instead of stretching the card. Its Quickstart button is Claira green rather than purple.
The chat list filters with the funnel icon used across Claira. In the sidebar and in the full Chats view, the filter is one funnel button that turns green with a count while a filter is on, lists Me with your name beside it, and ends with Clear filters. It replaces the labelled Filter by user button in the Chats view. See Filter the list by teammate.
The Chats view is tidier. Active and Archived sit in a view menu beside the search box, a button clears your search while you have one, and the count at the bottom reads 1 chat or 4 chats. See Search every chat.
The chat you have open is highlighted in Claira green instead of blue, and so is the place a pinned chat will land while you drag it.
Every icon button in Agent Mode explains itself. The attach button beside the message box, the Show history and New chat buttons in the top bar, a chat's actions button and the remove button on an attached file now show a tooltip, and the button that folds the sidebar away is announced as Hide history. Long chat titles, the marker on a branched chat, the reordering hint on Pinned, attached file names and the notice on a teammate's read-only chat show their full text on hover and on keyboard focus, not on hover alone.
The chat's document count reads correctly. Hovering the paperclip in the top bar now says 1 document in this chat or 4 documents in this chat, instead of "document(s)". See Attach documents to a chat.
The chat filter also filters by date. In the sidebar and in the full Chats view, the filter now has a Date section (Any time, Last 7 days, Last 30 days, Last 90 days, counted from each chat's last activity), the funnel appears as soon as the case has a chat, and a list the filters leave empty says No chats match your filters. See Filter the list by teammate.

Multi-Code: the card reads top to bottom, and every control says what it does

The card now reads in the order you use it. View History and Quickstart sit in the first row, then the shared instruction, your field rows, Add field and the scan buttons. The separate Insert into selector is gone: when you insert from history or Quickstart, Claira asks where the text goes and starts on the destination you chose last time. See Multi-Code mode.
The shared instruction opens and closes. Its arrow turns as it opens, and while it is closed the first line of your instruction shows beside the title, so you can tell at a glance whether one is set. Claira remembers in this browser whether you left it open.
Each row's Destination field uses the same picker as single and bulk scans, with the used and pending markers, long names shortened with the full name on hover, and Create new field, which puts the new field straight into that row. A field already chosen in another row shows Used in another row and cannot be picked again, because two rows writing to one field would overwrite each other. Screen readers now announce each row's picker and prompt box together with the row's number.
Add field is a full-width row under your fields that shows how many of the 10 you have used. At 10 it is disabled and explains why when you hover over it, instead of a warning appearing after the click.
Scan Document and Start bulk scan sit side by side. Start bulk scan stays visible, disabled with the tooltip Select one or more documents to get started until you select documents, and no longer carries a count: the number you have selected shows in the card's header, which now has the green mode tile and a help button.
Reset fields asks first and says what it clears; values already written to your documents do not change. Each row's move up, move down and remove buttons are named and show a tooltip, and the first row cannot move up nor the last row move down.
Problems in a response read as plain notices. When a single-document response has a value for a field you did not configure, or a row has no destination field, the card shows a notice in your language that names the field, instead of an English line starting with [error] or [warning].
Prompt Lab and Bulk Scanning share the same entry row: Save Prompt, View History and Quickstart, which turn into icons with tooltips in a narrow panel. Quickstart is Claira green rather than purple, the Prompt Lab bulk button also reads Start bulk scan, and the Prompt Lab header shows your selection only when more than one document is selected.

Your prompt and your Multi-Code setup save as you type

The prompt in Prompt Lab and Bulk Scanning saves on its own. Beside Save Prompt, Claira shows Unsaved changes while you type, Saving your changes a moment after you stop, then All changes saved. Save Prompt still adds the prompt to your prompt history. Clearing the box saves an empty prompt, so the box is still empty after a reload. See Saving and resetting.
Your Multi-Code setup saves the same way: the shared instruction, every row (destination, label, instruction and order), the template, a conversion from Prompt Lab, a Quickstart insert, and Reset fields once you confirm it. The status shows in the first row of the Multi-Code card. See Multi-Code mode.
Nothing is saved behind your back any more. Switching modes used to save everything on screen, including model settings you had changed in Settings without saving them. A mode switch now saves only the mode, after sending any change that is still on its way, and a waiting change is also sent when you switch browser tabs or close Nuix Discover.
Your work comes back as you left it after a reload or on another device: the prompt and the Multi-Code setup, along with your Scan as choice, how you last ran a bulk scan, and your default chat visibility.
A save that fails says so. Your text stays on screen, and the status reads Your changes were not saved with the reason: Claira could not be reached, did not accept the change, or could not read the case, or the text is too long (more than 100,000 characters, or 20,000 for one Multi-Code field instruction). Try again retries, and so does your next edit.
Leaving Local Mode confirms only a real save. Settings saved successfully now appears once the model has actually been saved; otherwise Claira says Could not save settings and gives the reason.

Case Context: decisions explained in Claira, template tiles, and one way out of a context

Approving, reverting, deleting and removing a section that has text now ask in Claira, in a dialog that lists what will happen, instead of a browser pop-up. The Draft button is now Revert to draft, and its dialog says that prompts already generated from the context do not change. A deleted context is confirmed with a message. See Approval.
A case with no contexts shows the five templates as tiles. Click one to open New context with that template chosen and the name filled in. See Creating a context.
The default only shows once a case has two contexts or more: a Default star on its card and a star button in each context's header, with a tooltip saying what the default is for. An approved context can be made the default. See The default context.
The header shows your live selection (3 documents selected) and an Approved or status label, while the context's own document count has a line of its own. Hover the icon beside the name to see which template the context came from; the coloured template label is gone.
Approving saves your unsaved edits first and clears the follow-up questions at once. Add to context is not shown on an approved context, and a note explains why.
Generating from your selection saves your edits first and pauses editing until it finishes, and screen readers announce it. The buttons no longer carry counts, the button on an empty context reads Generate from selection, and with nothing selected Claira asks you, in Case Context's own words, to select one or more documents to add. See Generating content from documents.
Errors are specific. A failed generation, deletion, default change or question dismissal now says what went wrong in your language, and a dismissal that fails is reported instead of failing silently.
One way out of a context. All contexts above the name and Close at the bottom both take you back to the list of contexts; Close used to send you to Prompt Lab. Both wait while Claira is loading, saving or generating, and neither keeps changes you have not saved. See Saving your work.

Settings: two groups you can fold away, model settings that save themselves, and one-click log fields

Settings is split into Your settings and Team settings, each with a badge saying who it affects: Only you, in this case or Shared with everyone in this case. Every section except Language opens and closes, only Connected fields is open on your first visit, and your browser remembers the rest. A closed section shows a one-line summary, such as the model in use, how many fields are connected, whether overages are on, or a scheduled deactivation date. Language closes Your settings, with the note Applies to all your cases. See First-Time Orientation.
Advanced settings is now Model settings, and it saves as you go. The Save button is gone: the bar at the bottom of Settings shows the save status, the reason and Try again if a change fails, and Done to return to Prompt Lab. The Settings Scope paragraph, the Last loaded time and the footer Refresh button are gone too. See Saving and resetting.
Reset to default is now Reset model settings, and it resets only the model, temperature, Top P and reasoning level. It no longer wipes your prompt, your Prompt Log and Log Field mapping, your default fields or your Multi-Code setup, and it is unavailable while everything is already at its default. The help buttons beside each control now announce what they explain rather than the control's current value.
Local Mode and Bring Your Own Model open in their own windows, from Local Mode settings and Connect your own model (or Edit your model connection) in Model settings. Choosing Bring Your Own Model in the model list before a connection exists opens its window instead of the help article: closing it keeps your current model, and saving a new connection selects it. A pencil beside the list reopens the connection, a note says it is shared with everyone in the case, and revoking it asks first. Leave Local Mode opens the Local Mode window on the question of which model to use next, and closing the window never stops a running bridge. See Bring Your Own Model and Leaving Local Mode.
Background processing shows one Connected account row once a token is connected, with the account name and when it was last verified. An options menu offers Replace token and Revoke (which asks first), token errors are explained in your language, and the section links to the token step of the setup guide. When background processing is not set up, the bulk scan confirmation now tells you to open Background processing in Settings and connect an account with a personal API token. See Setting Up Background Processing.
Manage plan always shows your plan and an Upgrade button, even while it is closed. With one larger plan, Upgrade goes straight to its confirmation; with several, it opens the plan list; it is hidden on the top plan and while a deactivation is scheduled. Scheduling or cancelling a downgrade now asks in Claira. See Manage plan after activation.
Email preferences show one row per email type with a dropdown on the right, stacked in a narrow panel. See Email Preferences.
Connected fields fit the panel. Long field names are shortened with the full name on hover, the Edit and Disconnect buttons name the field they act on, and disconnecting asks first in Claira. Change model, Open plan manager, the Agent's View plans and Field Migration's credential notice now open the right section of Settings and scroll to it. See Fields.
New: one-click Log Field and Prompt Log field. While either is missing, Connected fields offers Create and connect. Claira gives the role to a connected Memo field that already carries the name, reconnects a matching field the case used before and has since disconnected, or creates a Memo field named Claira Log or Claira Prompt Log in your language, with a number if the name is taken. It works one field at a time, reports each result (screen readers announce them), and offers access instructions to copy for your administrator. If Discover already holds the name for a field Claira has never had connected, Connect new field opens with the name and the role filled in, so only the composite ID is left to add. See Setting up the Log Field and the Prompt Log field.
The Canadian flag beside Canada-hosted models shows the official eleven-point maple leaf.

Insights are free

Generating and revising an Insight no longer uses tokens, from the Insights page, from an Agent Mode card, and in Local Mode, whatever the number of source documents. The token cost table lists bulk scans only, and Insights says so. Insight charges made before this release stay in the billing ledger under their original label.
Agent Mode now says plainly what costs tokens: only a bulk scan you approve on a card, charged like any bulk scan. Chatting and Insights are free, and Agent Mode never depended on your token balance.
For organization admins: with role enforcement on, Claira now also checks the Create an Insight and Revise an Insight permissions, on the Insights page, in Local Mode and on Agent Mode cards, and the Roles screen no longer marks them Not yet enforced. Both are part of the built-in User role, so only custom roles without them are affected. See Users and Roles.
For organization admins: with role enforcement on, Claira now also checks the Rename an Insight and Regenerate an Insight title with AI permissions, on the Insights page, in Local Mode and when Claira renames an Insight in Agent Mode, and the Roles screen no longer marks them Not yet enforced. Both are part of the built-in User role, so only custom roles without them are affected; anyone else sees a message saying their role does not allow it. See Users and Roles.

Prompt History counts scans

The use count on a prompt now counts scans. Every single-document scan, bulk scan and Multi-Code scan that used the wording adds one, repeats included, and so does a bulk scan you approve on an Agent Mode card. Saving or inserting a prompt keeps it in history but no longer counts as a use. Hover the count to see how many were single scans and how many were bulk scans. Entries recorded before this release show as Earlier uses, which can include saves and inserts, so the count on an older prompt may be lower than it was. Prompt details now say how a prompt was used, or Not used in a scan yet. See How uses are counted.
Multi-Code configurations are kept in Prompt History. Each Scan Document and Start bulk scan in Multi-Code records the shared instruction and every field row, with its destination field and prompt. Opened from Multi-Code, the history starts with the new Multi-Code history filter on; turn it off to see every prompt. A configuration card shows a Multi-Code badge, its number of fields and its destination fields, and its details show the whole configuration. Restoring a whole configuration is not available yet, so a configuration has no Restore button. See Multi-Code history and Multi-Code mode.
The history export has five new columns: Kind, Fields, Single scans, Bulk scans and Earlier uses. The summary sheet says when the Multi-Code history filter was on, and the prompt PDF's usage line lists the same breakdown. See Exporting the whole prompt history.

Set a logging field from Agent Mode

Ask Claira to make a field your Log Field or Prompt Log field. In Agent Mode, Claira shows a Set a logging role card naming the field, the role and the field that holds the role now, and nothing changes until you select Set role. The role moves for everyone on the case from their next scan, values already written stay as they are, and a bulk scan already running keeps its Log Field. The card accepts connected Memo fields that Claira can write to, and Claira cannot connect an existing Discover field for you. See Giving a field a logging role in a conversation and Creating fields with Claira.
Every change to the Log Field or the Prompt Log field is now recorded in the case's audit trail, whether it is made in Settings, by the one-click setup or on an Agent Mode card. See Setting up the Log Field and the Prompt Log field.
The one-click log field setup stops as soon as your organization role refuses a step, now also when the step would create a field. The fields it did not try are listed as such, and the box no longer offers Try again, which would be refused the same way. See Setting up the Log Field and the Prompt Log field.
For organization admins: with role enforcement on, changing the Log Field or the Prompt Log field now requires the Designate the prompt or task log field permission, in Settings, in the one-click setup and on Agent Mode cards, and the Roles screen no longer marks it Not yet enforced. The built-in User role does not include it, so once enforcement is on, only organization admins and custom roles that grant it can move these roles; anyone else sees a message saying their role does not allow it. See Users and Roles.

Case Context saves as you type, captures more, and can be revised with the Agent

Case Context now saves your edits as you type. Approve takes the place of the Save button, and a status beside it tells you whether everything is saved. Long editing sessions no longer lose the edit lock, and if the context changes somewhere else while you type, Claira asks before overwriting anything. See Setting up and Troubleshooting.
Each section now shows its Instructions for Claira, the checklist Claira follows when it generates from documents. Templates fill them in and you can edit them on a draft. See Setting up.
Generating from documents now captures every material fact: names and their variants, roles, dates, amounts, identifiers and how people are connected.
Follow-up questions are now one Missing information list of up to five questions, each naming the section it completes, shown without stray quotation marks. Questions stay on the list until the documents or the context answer them. See The same questions keep coming back.
Revise with Agent on a draft opens an Agent chat about that context; a chat you used today opens again. See Revising with the Agent.
Smart Revise is now called Revise with Agent, and so is the revise panel of an Insight. Both work as before and offer Open Agent chat. See Rate responses and revise with Agent and Insights.

Reporting a problem

New article: Reporting a Problem lists what to send support when something goes wrong (a screenshot or screen recording, the document or its text, the prompt, where you were in Claira, the exact error message, and when it happened) and explains why: Claira doesn't log, track, or store the content of your documents, so we can't look it up ourselves.
The Troubleshooting page now links to it.

New Case Context template for arbitration

Arbitration now has its own Case Context template. When you create a context, pick Arbitration to start an Arbitration Context with four sections built for arbitral proceedings: Parties & tribunal, Background & procedure, Claims & issues, and Production & privilege rules. See Templates set the terminology.
Built from your documents, it follows how arbitrations run. Claira records the tribunal, the institution and rules, the seat and governing law, and the rules for document production. If the request for arbitration or the procedural order on document production is missing, Claira asks for it.
The Litigation / eDiscovery template's description now refers to court proceedings only, so it is clear which template to pick.

New help article: how we choose AI providers

A new article, How We Choose AI Providers, sets out the standards every AI provider must meet before its models appear in Claira: privacy (no training on your content, no logging or keeping of it), data residency, credibility, independent security audits, benchmark-tested quality, fitness for legal work, reliability, and no lock-in. It also explains how providers are approved and re-reviewed, and how we vet the software Claira is built on. Every model Claira provides meets all of them; models you connect yourself through Bring Your Own Model or Local Mode are your choice and are not covered.

Privacy and Security now covers Germany and states the AI model provider policies

Privacy and Security now lists all three deployments (Canada, Australia and Germany, the last at Frankfurt, https://nuix-de.claira.to) and how each one is kept inside its country.
Plain policies for every AI model provider: requests stay in your deployment's country, and providers may not train on, keep, or share your content. Contact support for the providers that serve your region. See also How We Choose AI Providers.
Encryption details are corrected: every deployment, including Canada, now uses customer-managed encryption keys (CMEK) for its database and media scan staging, held in that deployment's region.
Model Selection now describes where each model runs without naming the companies that host it, and points to the provider policies and to support for regional details.

One name for GPT-OSS 120B

In Australia, the model picker now lists GPT-OSS 120B, the same name Canada uses, instead of GPT-OSS 120B (SCX). See Model Selection.

Error messages show a reference code

When something fails, the error now shows a reference code. A line under the message reads Reference: followed by a code such as 7K4Q-92XD, with a copy button. Send that code to support and we can find exactly what went wrong, without needing your document. See Reporting a Problem.
Claira now tells you when the AI service is busy. If the AI provider is at capacity, Quickstart, Insights, Revise with Agent, Agent Mode and Case Context say so and ask you to try again in a minute. They used to show a generic failure, such as "Failed to generate prompt".

Admin Portal: manage cases, billing reports and email defaults

Organization admins can change a case's plan from its page in the Admin Portal, including card-paid cases through Stripe, and cancel a scheduled change. See Changing a case's plan.
They can deactivate a case at the end of its billing period and, once it is deactivated, schedule the deletion of its data with the same typed confirmation and 7-day grace period as in Claira. The case team and the admin are emailed. See Deactivating a case, Deleting a deactivated case's data and Deleting a case's data.
Each case page lists its billing reports by cycle: the same Billing Report as in Claira, with amounts, still available after a case's data is deleted. See Billing reports.
A monthly-invoice case can be marked as a demo or test case to keep it off your invoice. A reason is required, its tokens and overage are still counted, and card-paid cases cannot be marked. See Demo or test cases.
Email routing: set which audience new cases start with, and apply an audience to active cases in one confirmed step. The person activating Claira still decides the activation email. See Email Routing in the Admin Portal.
The Overview now opens on what needs attention, then shows active cases, active users, tokens this month, the open invoice and deactivated cases. See Overview.
Clearer screens: a Docs button on every page that opens the help site, filters behind one button, the sidebar stays in place on long pages, empty values read "None", and amounts show in your region's currency. See Finding your way around and Activity Feed in the Admin Portal.
In Claira, the Email preferences section explains that your organization's admins can set these too. See Email Preferences.

Billing Report: date ranges in words, and "None" where a figure is missing

The Billing Report PDF writes its date ranges with "to" and names what it does not have. The cycle dates under the title are no longer joined with a dash, and a figure the report does not have, such as the plan tokens in the all-time report or a blank case name, reads None instead of a dash or a blank. Claira and the Admin Portal now build this report from the same code. See Billing Report PDF.

Claira fits a pane as narrow as 320 pixels, and its small text is easier to read

Claira's modes and dialogs now fit a pane as narrow as 320 pixels: related buttons stack together instead of spilling over, and nothing scrolls sideways. Below 320 pixels, Claira asks you to widen the pane. First-time orientation
Small grey text is darker and easier to read, and rows show less of it: details such as who ran a task, when, and how far it got now share one line, and a field's technical ID appears when you point at the field name in a task's details.
Confirmation messages keep a comfortable width on a wide pane instead of stretching across it.

Image scans read TIFF and JPEG 2000 scans, and PDFs with extra data at the start

Image scan now reads TIFF and JPEG 2000 images. Nuix Discover records a file's real format when it is ingested, so a scanned or handwritten document saved as .pdf is sometimes a TIFF or JPEG 2000 image in the case, and an Image scan used to stop with "not a supported Image format". Claira now converts these images before the scan, and a multi-page TIFF is scanned as one document. See Media scans.
PDFs with extra data before the PDF header, which some scanners and scanning apps add, are now read like any other PDF.
Clearer messages when a file still can't be read. The message names the file type Discover lists for the document, and when the document has no extracted text either, Claira says so instead of suggesting a Text scan. See Image scan says the file is not a supported format.

A tidier Agent Mode message box

The Agent Mode message box now starts one line tall, with the paperclip, your text and the send button lined up in a row. It grows as you type, and once your message runs past two lines the paperclip and send button move to the bottom of the box, next to the line you're typing. See Agent Mode.

Agent Mode shows which documents each question was sent with, and PDFs attach again

Every question now shows the documents it was sent with. Just above each message you send, Agent Mode lists the Discover document you had open, by its title, and any file you attached. When you move to another document and keep asking, the next questions show the new one, so you can see where a conversation moved from one document to the next. Hover over a document to see what Claira received of it with that question: its text, its own file, or both. See Attach documents to a chat.
Picking a file now visibly does something. A file you attach appears in the message box straight away, with a spinner while Claira reads it, and waits there for your next message; you can remove it before you send. Previously the only sign was a small count in the top bar.
PDFs attach again. Claira reads the text of a PDF in your browser once more, and a scanned PDF or an image goes to Claira's secure storage as intended; before this fix, attaching a PDF failed with an error.
The document list in the top bar says when each document came in. Each entry shows the date of the question that first brought it in (click it to jump to that question), or Not sent yet for a file waiting for your next message.

Longer questions in Agent Mode, with a limit you can see

A chat message can hold up to 100,000 characters, about a long document's worth of text. As you approach the limit the message box shows a character count; past it, it says how many characters to remove and keeps Send unavailable until the message fits. Your text stays in the box while you trim it. Previously an over-long message was accepted by the browser and then refused by the server with a generic error. See Agent Mode.

Cancelling a scheduled change now emails only you

A cancellation is confirmed to the person who cancelled it, and to nobody else. Cancel a scheduled plan change, a scheduled deactivation or a scheduled data deletion, and Claira emails you alone. The rest of the case team is not notified, because the change they were told about is simply not happening. This is the same whether you cancel in Claira inside Nuix Discover or in the Admin Portal. See Email Notifications Overview.
The email names the case, what was cancelled and the date that no longer applies, arrives in your language, and ends with a note that it was sent only to you. If Claira has no email address on file for you, the cancellation still goes through, it is just not confirmed by email.
Booking is unchanged. Scheduling a plan change, a deactivation or a deletion still notifies the case's usual recipients. See Cases in the Admin Portal and Deleting a case's data.

Clearer wording on bulk task status

Bulk task status lines now read as plain sentences. A cancelled task says "Cancelled by user." followed by how many documents were written, and an expired one says "Expired: reimbursed", in place of the dash-joined phrasing used before. Tasks that finished earlier keep the wording they were written with; nothing was rewritten after the fact.

"View history" reads like the rest of Claira

The button that opens Prompt History is now View history rather than "View History", matching the sentence case used everywhere else in Claira. It is the same button in the same places: Prompt Lab, Multi-Code and Bulk Scanning. See Prompt History.

A clearer Agent Mode chat window

The Agent Mode chat window is now solid white with a clear outline, and the conversation sits on a light grey background, so you can see where the chat begins and ends. Your messages, Claira's answers and the message box stand out against it. See Agent Mode.

Lists in Insight PDFs line up with their items

Every bullet and number in a downloaded Insight PDF now sits beside its own item. They used to be drawn a line too low: the first item in a list had no bullet, each bullet appeared beside the item after its own, a stray bullet hung under the last item, and a numbered list gave each item the number of the one before it. An item long enough to wrap now reads as one item, with its second line indented under the first. See Downloading the PDF.
Numbered lists of ten items or more stay aligned. The numbers line up on the right and no longer run into the text beside them.

Even word spacing in Insight reports and the prompt history export

Words in a downloaded Insight PDF and in the full prompt history export keep their spacing. Some words used to be printed a little too close to the next one, so "P.C., M.P." could read as "P.C.,M.P." and a name could touch the citation after it. See Downloading the PDF and Prompt History.

Insights read your fields as they are when you click Generate

Insights now read the field itself, in Nuix Discover, at the moment you click Generate Insight. Before, an Insight over selected documents only found values a bulk task had written, so values written from Prompt Lab, by Multi-Code, by Field Migration or by hand were missed, and Claira asked you to select documents with values. Every value in the field now counts, however it got there. See Insights.
Corrections count. An Insight over a bulk task used to read the answers as they were when the task ran. It now reads the task's field as it stands today, so a value a reviewer corrected is read as corrected, and a value that was cleared is left out.
Your access applies. Claira reads the field with your own Nuix Discover access, so an Insight only reads values you can see.
Revisions build on the same evidence. A revision works from the values the Insight was first generated from; to pick up later changes, generate a new Insight. Insights created before this release keep the values they were generated from.
If Claira cannot read the field, it now says so and creates nothing, instead of suggesting that your documents have no values.

Watch an Insight being written

The Insight viewer now shows what Claira is doing while an Insight or a revision generates, instead of a spinner. Four steps are ticked as Claira passes them, with a timer on the current one: reading the AI results (with how many it read), checking Document IDs, writing the Insight and checking references. See Watching an Insight being written.
The text appears as Claira writes it, a few words at a time, marked as a draft. Its citations stay grey until the finished Insight replaces the draft, then turn green and link to their sources as before. The draft is not kept: only the finished Insight is saved.
Grey placeholders show what is coming. Before the first words arrive, grey lines show where the text will appear; while Claira writes, one grey line under the draft shows more is coming and a grey bar holds the title's place. The finished Insight fades in when it replaces the draft.
An interrupted Insight no longer generates forever. If generation stops partway, after about 10 minutes the Insight says it stopped and nothing was saved, so you can try again. An earlier revision is unchanged.

Watch a Case Context fill in from your documents

Generating a context from documents now shows what Claira is doing. When you click Generate from selection or Add to context, three steps are ticked as Claira passes them: fetching the documents (with a count such as "2 of 5"), reading them (with a timer, since this can take a few minutes on a large selection) and updating the sections. See While Claira generates the context.
The sections show where the result will land. Empty sections show grey lines, and text that may be replaced sits under a soft moving highlight. When the result arrives, the sections Claira changed fill in one after another with a brief green outline; the others stay as they were. With reduced motion turned on, the text appears at once.
Your other work is left alone. If you open another context before the generation finishes, the one on screen is not touched and a message names the context that was updated. While a generation runs, the Agent does not edit that context; it asks you to try again once it finishes.

Multi-Code fills up to 10 fields

Multi-Code now fills up to 10 fields in one pass, up from 8, with every model. Add field counts to 10, Use Multi-Code can turn a prompt into as many as 10 field prompts, and Revise with Agent can suggest a change to every field and to the shared instruction at once. See Multi-Code mode.
The price does not change. A bulk Multi-Code scan still costs the document type's base rate plus 1 token per document, whether you use 1 field or 10, and Multi-Code on a single document is still free. See Token usage.

Change the field in one place and the other follows. Picking a destination field in Prompt Lab now changes the one in Bulk Scanning straight away, and the reverse, including a change made in the bulk scan confirmation or a destination you approve on an Agent Mode bulk scan card. Before, only your prompt carried over: a bulk scan started from Prompt Lab could write to whichever field Bulk Scanning had been left on, with nothing to show it. See The field Claira starts you on.
Start Bulk Scan on # Documents in Prompt Lab writes where Prompt Lab shows. The confirmation it opens names that field, so you can check it before you start.
If the two modes remembered different fields, both now open on the field you last used in Prompt Lab.
A scan that is already running is not affected. It keeps writing to the field it started with, and resuming a task writes to that task's field without changing your pickers.

Image scans on three more models (Australia and Germany)

Llama 4 Maverick and Gemma 4 31B (Australia) and Qwen 3.8 Max (Germany) can now run Image scans of PNG, JPEG and WebP images and of single-page TIFF or JPEG 2000 scans. Until now, media scans needed a Gemini model. PDFs, multi-page scans, audio and video still do. See Which models can run a media scan.
With one of those models selected, Auto sends a document's image when the model can read it, and its extracted text otherwise, so a PDF is reviewed from its text rather than refused. If you choose Image and a document's file is a format the model does not read, the message names the file type and offers Change model.
The other models review text only. Model Selection now says for each model which scans it runs, and the message shown when a model can't run a media scan now reads "The selected model can't scan this type of media" instead of saying every media scan needs a Gemini model.

Missing information asks only for what your team knows

The Missing information list below a Case Context now asks only for what your team knows and the documents do not say: for example which party you represent, the document requests or order the review answers, the notice that sets a regulator's mandate, or the warrant a law enforcement review works under. It no longer asks for email addresses, job titles or other details that change no review decision, or for conclusions the review itself is meant to reach. See Generating content from documents.
Each question now names the parties, documents or dates it is about, so you can answer it without opening the documents.
Claira no longer takes the party you represent from whose documents are being reviewed. It keeps asking until a section of the context says it. See The same questions keep coming back.

Organization admins can now open the Admin Portal without going through Nuix Discover. Turn on email sign-in for yourself under Settings > Email sign-in: enter your address and type the six-digit code Claira emails to it. From then on, go to app.claira.to, enter that address and choose Email me a sign-in link, then open the link and choose Sign in. The link works once, for 15 minutes. See Opening the portal and Email sign-in.
Your organization stays in control. Email sign-in is on for the organization by default and each admin decides for themselves. Any admin can turn it off for everyone, which removes every admin's email sign-in and ends sessions opened with an email link. Settings lists which admins use it; each admin's address is visible to them alone.
The portal's front page answers Check your inbox for any address, so it never reveals who uses email sign-in, and it remembers your region after your first visit. The Verified card now shows whether you signed in from Claira in Discover or with an email link.

See, and choose, whether the open document goes with your message

The message box now shows the document you have open. It's the first chip above your text, outlined in green with its document ID and title, and it says what Claira will receive of it with your next message: Text, which switches to Text and file as you type once your message refers to the document, since that is when Claira reads the file too. Until now the open document's text went with every message without anything on screen saying so. See The document you have open.
Keep it out of your messages with one click. The × on the chip stops the document going with your messages, text and file alike, and Claira is told a document was held back instead of guessing at it. The chip stays, marked Not sent, until you send it again or open another document. See Keep the open document out of your messages.

Prompt History marks the prompts an AI wrote

A small wand now marks every prompt one of Claira's AI tools wrote. Hover it to see which: Suggested by Revise with Agent, Written by the Agent (a prompt inserted from Agent Mode, or approved on an Agent card) or Written by the Prompt Generator. The wand appears on the card and on the version row, in both views, and prompt details say the same in words. Templates and prompts you typed get no wand. See Prompts written by AI.
Applying a Revise with Agent suggestion now saves it to Prompt History straight away, so every revision the AI proposed and you accepted is on record, even before you scan with it. Its details offer Open Agent chat to reopen the chat the revision was recorded in; only you see that link, since the chat is yours. See Rate responses and revise with Agent.
The prompt history export gains an Origin column: Revise with Agent, Agent, Prompt Generator or Template, blank when Claira has no record of who wrote the revision. See Exporting the whole prompt history.

Tables in Insights, and Insights in Excel

Insights can now hold tables. Ask for one in your prompt, naming the columns you want, and Claira writes a table; it also chooses one on its own when the answer is a set of like records, such as a chronology or a count per custodian. Every row keeps its citations, and dates are written so the table sorts. In a narrow panel each row shows as a card. See Tables in an Insight.
Switch between a table and text in one click. Beside Revise with Agent, Rewrite as text or Rewrite as a table fills in the request for the other format; edit it if you like, then select Generate revision. Every fact and citation carries over, and the earlier revision stays in the sidebar.
Download an Insight to Excel. Download Insight now offers PDF (.pdf) or Excel (.xlsx). The workbook has a summary, one sheet per table with real dates you can sort and filter, the whole report, and a Source documents sheet listing every document the Insight read, with its title, whether it is cited and the value it was generated from. See What the Excel workbook contains.
Tables print as tables in the PDF, with their header row repeated at the top of every page they run onto.
The Chronology template now produces a dated table (Date, Event, Parties, Sources), with gaps listed under it. See Chronology.

Search term families, ready to import into Nuix

Ask the Agent for search terms and get a file Nuix imports as it is. Claira writes a search term family for Nuix Discover, with one term per concept, each a short label and a query in Nuix's search syntax, and checks every query before showing it. The card's Download spreadsheet button saves an Excel file with no header row (label in column A, query in column B) for Add search terms > Add terms from a file. You no longer need to paste a syntax instruction block into the chat. See Generating Search Term Families.
Start from a Case Context. Generate search terms in a context's header opens a new Agent chat that asks for a family built from that context, on drafts and approved contexts alike. See Setting up Case Context.
The card stays with the chat, so you can reopen the chat later and download the file again. It is free.

A getting-started email the first time you open Claira

The first time someone opens Claira in an activated case, they now receive a short getting-started email in the language they use Claira in. Each person receives it once. It is not sent to people who already used Claira, to anyone who received the case activation email in the last 14 days, or to anyone who unsubscribed from Claira's marketing emails. See Getting-started email.
Privacy and Security now says where email is processed. Claira's emails are delivered by providers in the United States in every region, the one exception to in-country processing, and the article lists what they receive. The emails Claira sends to your team never contain document content or AI answers. See Email.

Case Context reads scanned documents, recordings and videos

Case Context now reads each document's own file, not only its extracted text. When you build a context from selected documents, Claira sends each document's PDF, image, audio or video file alongside its extracted text when it fits, so handwriting, stamps, signatures, tables and poor OCR are read from the file itself. See What Claira reads from each document.
Documents without extracted text can feed a context. A scan without OCR, a recording or a video no longer stops the run with "Please select a document with extracted text": Claira reads it from its file.
You can see how each source was read. Under Source documents linked to Case Context, and in the exported PDF, each document now says Read from extracted text, Read from native file, or Read from native file and extracted text.
A document Claira cannot read is left out, and named. When a document has no extracted text and its file cannot be sent (larger than 15 MB, too large alongside the other files, or not downloadable), Claira builds the context from the others and tells you which documents it left out. See A document was left out, or read from its text only.
In Local Mode, Case Context still reads extracted text only.

Organization prompt templates in Quickstart

Share your organization's own prompt templates with everyone. Organization admins now write prompt templates on a new Templates screen in the Admin Portal: a title, an optional description, a category, the language the prompt is written in, and the prompt itself, with a live preview of how it will look in Quickstart. A template can be archived to take it out of Quickstart, restored, or deleted for good, and every change appears in Activity. See Prompt Templates in the Admin Portal.
Quickstart shows them first. In Prompt Lab, Multi-Code, Bulk Tasks and the Agent's Bulk scan proposal card, a Your organization section now sits above Claira's own library whenever your organization has templates. Templates in your interface language come first, and one written in another language carries a small EN, FR or DE tag. Insert one exactly like a library template. See Prompt templates.
The Use prompt templates permission is now enforced. With permission enforcement switched on, it decides who sees the Your organization section; Claira's own library stays available to everyone. Claira now checks ten permissions. See Users and Roles in the Admin Portal.

Give permissions in Agent Mode

Choose, for each chat, whether the Agent asks before running. A small control under the message box sets the chat to Ask before running (how Agent Mode has always worked: scans, Insights and Case Context edits wait for your approval) or Give permissions. See Ask before running or give permissions.
In Give permissions, the Agent sets up and starts the work. It writes the prompt, uses your current selection, chooses the destination field and tells you what will run; the bulk scan or Insight card then starts after a 5-second countdown you can stop with Don't start. A bulk scan started this way costs the same tokens as one you approve yourself, and Case Context draft edits are saved without a confirmation step.
Some things always wait for your click. Scanning every document in the case, creating a field and giving a field a logging role still need you to press the button on the card.
Admins decide what is allowed. A new Agent section on the Admin Portal's Settings screen sets Give permissions to Not allowed, Allowed, new chats ask first (the default) or Allowed, new chats give permissions. See Settings and Discover Portals.

Field Migration: long values fold down to a few lines

Long values no longer take over the screen. In the one-document test result, the pick-list matching table and the migration history, a long value now shows its first few lines with Show more to read the rest and Show less to fold it away. Only the display changes: nothing is cut from what is written. See Field Migration.
The matching table no longer cuts values to one line. A long source value used to be cut short on a single line, so two values that differed only near the end looked the same; it now wraps onto a second line before folding, and the choice beside it stays level with the value's first line.

Background processing for image, audio, video and Auto scans

Every scan type can now run in the background. On a case with Background Processing set up, the Run as background processing task checkbox in the bulk scan confirmation dialog is offered for Image, Audio, Video and Auto scans as well as Text, single-field and Multi-Code alike. See Running media scans in the background.
Claira uploads the files first. For an image, audio, video or Auto scan, Claira uploads each document's file from your browser before the scan starts on the servers. The bulk scan card shows Uploading files for Task #NNN with a count of files uploaded and a Pause upload button. Keep Claira open until the upload finishes; after that it is safe to close, and the completion email arrives as usual. See Image, audio, video and Auto scans upload their files first.
Anyone in the case can finish an interrupted upload. If Claira is closed or the upload is paused, the task waits under Ongoing background tasks as Uploading, and Finish uploading uploads the rest while keeping the files already uploaded. After 24 hours Claira starts the scan with the files it has; documents without a file are not scanned and not charged, and a task with no file at all is cancelled and fully refunded. See Uploading files for image, audio, video and Auto scans.
Send to background works for media scans too. Claira first uploads the files of the remaining documents, then the scan continues in the background. See Moving a live scan to the background.
Same price, same limits. A background media scan costs the same tokens as the same scan in your browser, the 50 MB per-file limit still applies, and a document whose file could not be used gets an E-FETCH note and is not charged. Image, audio, video and Auto scans the Agent starts still run in your browser. Uploaded files are deleted when the run ends, and after 14 days at most. See Privacy and Security.

Insights and Case Context download as Word documents

Download an Insight or a Case Context as a Word document. Download Insight now offers Word (.docx) beside PDF and Excel, and a Case Context downloads from Download Case Context as PDF (.pdf) or Word (.docx). The Word file is the same report as the PDF (header, status badge, every section, source documents, disclaimer), editable, to drop into your own memo, brief or chronology. Headings are real Word headings, and lists and tables are real Word lists and tables. See Downloading the Insight and Downloading a context.
Word keeps every character. Text the PDF font cannot show (Chinese or Cyrillic text, symbols such as ≥) comes through intact in Word. When a PDF has to replace characters with ¤, Claira's message after the download now suggests the Word version.
Citations link both ways in Word. In a Word Insight, a citation jumps to its source document, and each cited source links back to the place it is first cited.
Downloads are recorded. Every Insight (PDF, Word or Excel), Case Context and Bulk Task History Report download is now recorded for your organization's compliance records (who, which case, what, which format, when), as prompt exports already were. The record never holds the content of the file; if it cannot be saved, the download is cancelled and Claira asks you to try again.
A Case Context download waits for generation to finish, so the file never captures half-written sections.
The Exports article now lists everything you can download from Claira, and explains when to choose PDF or Word.

Early case assessment: a first look at the whole case

A new mode, Early case assessment, gives you a first look at a whole case before anyone has scanned anything: how many documents it holds, whose they are, what kinds of file they are, when they are dated and who writes to whom, with suggested review segments and next steps. Open it from More > Early case assessment, beside Case Context. See Early case assessment.
It reads the metadata of every document, with your own Discover access, however large the case: nothing is sampled. Claira counts the case's standard fields (custodians, file extensions, document types and kinds, languages, senders, recipients and authors) and up to five pick lists you add, and writes an overview from those counts. It never reads the documents' content, it writes nothing to Discover, and it is free. Reading takes about a minute for every 200,000 documents.
It shows the makeup of the case as a whole. File types are grouped into families from each document's extension and split into user-generated and system-generated files; a timeline marks the months where documents spike or thin out; tables break custodians down by year and by file type; custodian coverage shows when each custodian's documents start, stop and pause; Who writes to whom lists the most frequent sender and recipient pairs; and when the case has a file size field, the assessment counts empty and very large files. See How file types are grouped.
Every figure comes with its source. Under the overview, What it was built from shows all of these statistics, with the counts for each field, the document dates, the imports and ingestions, and a sample of titles.
Each run is a new, dated assessment, listed under Past assessments, and Download PDF exports the overview with the statistics on a page of their own. Claira keeps each assessment with the statistics it was built from until someone deletes it. See Privacy and Security.
Claira tells you when the case has changed. When imports have loaded documents since the latest assessment, the mode says how many and suggests running a new one.
The Agent can read the latest assessment when you ask about the case, for example how many documents it holds or who the custodians are. See Agent Mode.
Bulk scan proposals for the whole case now say how many documents it holds. When the Agent proposes a bulk scan of all documents, the proposal card shows All documents with the count from the latest early case assessment and its date, instead of how many documents you have selected. With no assessment yet, it suggests running one. See Agent Mode.
It replaces the Early case assessment Insight template, which wrote a memo from a bulk scan's document summaries. The template is no longer offered, Insights already generated from it are unchanged, and its help article now leads to the new one. See Insight templates.
Early case assessment uses a Claira-hosted model, so it is switched off while Local Mode is on; past assessments stay available. See Local Mode.

Bulk Task History and Billing Report files are named in your language

The Bulk Task History Report and the Billing Report now save under a name in your interface language, as Insights and Case Context already do: for example Claira-Bulk-Task-History-Report-Acme v. Byrne-20260921.pdf, in French Claira-Historique-des-tâches-en-lot-Acme v. Byrne-20260921.pdf, in German Claira-Verlauf-der-Massen-Aufgaben-Acme v. Byrne-20260921.pdf. The date at the end is the day you download it, written year, month, day so your files sort in order. See Bulk Task History Report.
A Billing Report's name says what it covers: the cycle for one month (Claira-Billing-Report-Acme v. Byrne-2026-08-20260921.pdf) or All-time for the whole case history, so two reports downloaded the same day no longer share a name. The Admin Portal names its billing reports the same way. See Billing Report.
A case with no name no longer produces a file called "Unknown"; the case part of the name is simply left out.
If one of these reports cannot be created, Claira says so in a full sentence and asks you to try again in a moment, instead of showing a technical error.

New Case Context template for employment disputes

Employment disputes now have their own Case Context template. When you create a context, pick Employment to start an Employment Context with four sections built for individual employment disputes: Employer, employee & representatives; Employment history & key events; Jurisdiction, claims & issues; and Privilege, confidentiality & sensitive data. See Templates set the terminology.
It works in any country. The Jurisdiction section records whose employment law applies: the country, the province, state or region where the law differs within it, and any federal, national or sector-specific regime, along with the forum. Claira fills it only from what your documents say, never from where a party has an office or the language of the documents, and asks your team when the documents do not settle it.
It follows how employment matters unfold. Claira records the terms of employment, performance and discipline, leave and accommodation, the termination and the reasons given for it, and material that needs care, such as medical information, personnel files and settlement discussions. If the claim, the complaint or the termination letter is missing, Claira asks for it.
For a union grievance, pick Arbitration; for an HR investigation, pick Internal investigation. The Employment tile says so.

Multi-Code bulk tasks no longer offer an Insight for a prompt they did not run

A finished Multi-Code bulk task no longer offers the wrong Insight. If the Prompt Lab box last held a prompt template when you started a Multi-Code bulk scan, the finished task could say "This task ran the prompt they read" and offer that template's Insight, and a new Insight could pick the task as that template's source, although the scan ran your Multi-Code fields' prompts. The offer now comes only from a bulk task that ran the template's prompt. A Multi-Code task started before this release may still show it; select Not now to dismiss it. See Insight templates.

Bulk scans approved in Agent Mode no longer offer an Insight for a prompt they did not run

A bulk task you approve in Agent Mode no longer offers the wrong Insight. If the Prompt Lab box held a prompt template when you selected Approve and start on a Bulk scan proposal card, the finished task could say "This task ran the prompt they read." and offer that template's Insight, and a new Insight could pick the task as that template's source, although the scan ran the prompt on the card. Your next scan in Bulk Scanning could do the same, because approving the card puts its prompt in the Prompt Lab box. The offer now comes only from a bulk task that ran the template's prompt. A card that runs the prompt already in the box, unchanged, still counts as that template. A task started this way before this release may still show the offer; select Not now to dismiss it. See Insight templates and Agent Mode.

A Team plan between Pro and Business, a 500-token pack, and token packs on your monthly invoice

New Team plan. Team sits between Pro and Business: $1,499 a month for 85,000 tokens, about 1.8¢ a token. Choose it from the plan manager in Settings or when you activate a case. See Plans and Billing.
A smaller token pack. Packs now start at 500 tokens, $50 on every plan.
Claira offers a pack when a bulk scan leaves you low. When a bulk scan finishes and leaves the case with less than 15% of its monthly tokens, or using overage, the message reporting the finished scan offers Buy tokens. The offer appears in Claira only, never by email. See When a bulk scan leaves you low.
Cases billed by monthly invoice can add a pack to the invoice. Buy tokens now offers Add to monthly invoice on these cases instead of a card payment: the tokens land straight away, and the pack is billed on your next invoice at $0.10 a token, with your partner discount applied. Organizations that enforce roles control who may do this with the new Add token packs to the monthly invoice permission. See Adding a pack to your monthly invoice.
The Admin Portal shows the packs. The partner invoice has a Token packs column, the case's billing reports list the packs in their amounts, and the activity feed records each pack with who added it. See Billing and invoices.

Help desk home page

The help desk home page now lists the same articles in the same order in English, French and German, including Insights, Agent Mode and its limitations, and the Translation workflow.

A template picked with Quickstart on an Agent card now counts as that template

A bulk scan you approve in Agent Mode now counts as the template you picked for it. If you picked a prompt template with Quickstart on a Bulk scan proposal card, the finished task offered nothing for it, and Prompt History marked its prompt Written by the Agent. The finished task now offers the template's Insight, as a task you start from the template yourself does, and its prompt has no wand. Editing the wording on the card keeps it that template; a prompt from the Prompt Generator, or one the Agent rewrites, is not one. Approving the card also puts the template in the Prompt Lab box with its prompt, so your next scan in Bulk Scanning counts as the template too. A task approved this way before this release keeps its wand and offers nothing. See Agent Mode and Prompts written by AI.

Working notes

Claira now keeps internal working notes about how people and teams use it (kinds of tasks, roles and how colleagues work together), to improve your experience. They never contain case or document content, are not shown or used yet, and are deleted with the case. See Privacy and Security.

The help desk sidebar now lists the same articles, in the same order, in English, French and German. In German it now shows every workflow, where it listed only Objective Coding, and the Case Context and Release Videos sections, which were missing. In French, Getting Started now includes What Claira can do for your review. In both languages, each section lists its articles in the English order.

Insights as a timeline, and a Format choice

Choose the format of a new Insight. The New insight card now has a Format choice: Automatic (as before, text or a table when the answer is a list of records), Text, Table or Timeline. The Chronology template chooses Timeline for you. See How to run an Insight.
See a dated Insight as a timeline. Any Insight whose table has a Date column can now be viewed as a timeline: events in date order, grouped by year, along a strip of bars that shows the whole period at a glance. An Insight asked for as a timeline opens on it; on others, View as timeline appears above the table, and the menu at the top right switches between Report and Timeline. See Seeing an Insight as a timeline.
Filter by party, search, and open the documents behind each event. Click a person or organisation on an event to show only the events that involve them, search by any word or Document ID, and expand an event to see each document's own AI result and the date it gives. Dates known only to the month or the year are marked as such, and undated events are listed last.
Claira flags dates the sources disagree on. When a document behind an event gives a different date, the event says Sources disagree on the date and the document is flagged.
Download the timeline as a PDF. Downloading the PDF from the Timeline view draws the timeline itself, with linked citations; Word and Excel keep the table. See Downloading the Insight.

Celebrations, milestones and While you were away

Small celebrations when work finishes. When a scan, an Insight or a migration finishes cleanly, Claira marks the moment with a small check, a short friendly line and, now and then, a milestone such as "Your 10th Insight." or "This case just passed 10,000 documents coded." Each one plays once, takes less than a second and never covers your document; runs with failures lead with what went wrong instead. If you prefer none of it, switch off Celebrations in Settings > Your settings: your results and notifications stay. See Celebrations.
A summary when a bulk scan finishes. Instead of an empty panel, a finished scan now leaves a card with its result ("2,908 documents coded"), the colored result bar, a suggested Insight when there is one, and View details. A short note also tells you when a running scan is halfway there and nearly done. See Bulk Scan.
While you were away. When a background task you started finishes while Claira is closed, a card at the top of Claira lists it the next time you open that case; when it finishes while Claira is open, in any mode, a notification tells you. See Bulk Scan.
Insight ready. When an Insight you started finishes while you are in Insights, a notification says it is ready to read, with an Open button. See Insights.

Multi-Code: start from a template, or reload an earlier pass

Quickstart in Multi-Code now opens with a Multi-Code folder of templates that fill the whole card at once, shared instruction and fields together: Objective coding (title, author, recipients, document type and date), Relevance and issues and Privilege and types. Choose Load into Multi-Code. See Multi-Code templates.
Rerun an earlier pass from Prompt History. A Multi-Code configuration in Prompt History now has Load into Multi-Code, which puts back the shared instruction and every field with its prompt and destination field. See Multi-Code history.
Destination fields are filled in for you when the names match. A loaded field keeps the destination it ran with while that field still exists, or takes the case field with exactly the same name. Any field left without one says Choose a destination field and is skipped until you pick one. Loading asks first when the card already has something in it, and never changes values already in Nuix Discover.
A stopped Multi-Code bulk task resumes with the fields and prompts it started with, even if you have changed the Multi-Code card since. See If a scan stops early.

Claira issues and emails your monthly invoice

Claira now issues your organization's monthly invoice and emails it. For organizations billed by monthly invoice, the invoice is issued automatically on the morning after the invoice window closes, once the window's overage is final. It is issued by Noticia Solutions Inc. (dba Claira), carries a number such as EDC-AU-00004, and is due 30 days after it is issued (Net 30). If your organization paid in advance, the invoice is paid from that account credit first and shows the balance due; demo or test cases are listed at no charge. See Invoices Claira issues and How monthly invoice billing works.
Everyone whose role can view billing receives it, in their own language, with a summary of what the bill is for and the invoice attached as a PDF and a CSV. Organization admins can stop a role from receiving it with the new Monthly invoice row on the Email screen, and invoice emails appear in the delivery log. See The monthly invoice email.
The Billing screen shows where each invoice stands. Pick a closed window and the new Invoice card reads Issued, Held, Nothing to bill or Not issued yet. An issued invoice shows its number, dates and total, a Download PDF button and who it was emailed to, and it can be sent again. See The invoice card.
An issued invoice never changes. If something in its window changes later, the Billing screen shows the current figures beside it and says they differ from the invoice as issued. See When figures change after the invoice is issued.
A new Billing details card on the Payment methods and settings tab holds your legal name, billing address, tax or VAT number and purchase order or reference, printed on your next invoice. Without a legal name and address the invoice is held until you add them. See Billing details and Held invoices.

Use any template with your Case Context

Every template in Quickstart, your organization's own included, now has Use with context (Multi-Code templates excepted). It fills the template in from your approved Case Context with the Prompt Generator and puts the result in your prompt box: the template's wording and output values stay as written, its placeholders (the case description, known counsel) are filled from your context, and any other facts the task needs are added in a short Matter details section. With one approved context Claira uses it straight away; with several, you pick one. See Use a template with your Case Context.
The privilege template no longer assumes Canadian law. It now asks for the jurisdiction, and Use with context fills it in when your Case Context states one. See Identify privilege types.
If the context you picked is reverted to draft while Quickstart is open, Claira now says so and asks you to pick another one, instead of using a context that is no longer approved.

A resumed bulk task keeps the prompt it started with

Resuming a task no longer changes its prompt or its scan type. When you resumed an Incomplete single-field bulk task, Claira scanned the remaining documents with whatever the Prompt Lab box and the Scan as dropdown held at that moment, not what the task started with, so a prompt you had tried since, or a different Scan as choice, could code the rest of the task differently. A resumed task now uses the prompt and Scan as setting it started with. If Claira cannot read that prompt, it says so and scans nothing: try again, or start a new bulk scan for the documents that are left. See If a scan stops early.
On your own model or in Local Mode, only a Text task resumes. A task started as an Image, Audio, Video or Auto scan now asks you to switch to a Claira model, or to leave Local Mode, before it resumes, because changing Scan as no longer changes what the task scans. See Bring Your Own Model.

A prompt the Prompt Generator wrote on an Agent card is now marked as the Prompt Generator's

Prompt History now credits the Prompt Generator for a prompt it wrote on an Agent card. When you approved a Bulk scan proposal card whose prompt the Prompt Generator wrote, Prompt History marked the scan's prompt Written by the Agent. That happened for a prompt you inserted from the Prompt Generator with Quickstart on the card, and for one the Agent had the Prompt Generator draft from your goal. That prompt now gets the Written by the Prompt Generator wand, even if you edited it on the card. A template you fill in with Use with context on the card gets this wand too, and the finished task still offers the template's Insight. A prompt the Agent writes or rewrites itself keeps Written by the Agent, and so does a prompt written on a card before this release. See Prompts written by AI.

Automatic model routing

Claira can answer with another model when yours is busy. Turn on Automatic model routing in Settings, under Model settings, beside the model picker. When the model you use is over capacity or not responding, even after Claira's usual retries, Claira answers with another approved model in your country instead of telling you the AI model is busy. It tries the other models in a fixed order for up to about a minute, and if every model is busy you see the same error as before. See Automatic model routing.
Your documents never leave your country. Claira only moves to models already offered in your model picker, held to the same model provider policies. See How We Choose AI Providers.
It is yours, for each case. It starts on in cases created from this release and off in cases that already existed, and each person can turn it on or off for each case.
It covers every AI feature: single scans, bulk tasks run in your browser, Multi-Code, Agent Mode, Case Context, Insights, Early case assessment, the Prompt Generator and Quickstart, Revise with Agent, and the names Claira gives your chats and bulk tasks. It does not apply to Bring Your Own Model, Local Mode or bulk tasks run with Background Processing, and it only moves to a model that can do the job: for example, an Image, Audio or Video scan only moves to a model that can read that file.
Nothing appears in the panel when Claira switches, and the record of each scan and chat reply keeps the model that actually answered. Organization admins see each switch in the Admin Portal's Activity screen under the new Model routing type. See Activity.
When a scan fails because the AI model is busy, E-AI now suggests turning on automatic model routing.

Bulk scans approved in Agent Mode run the prompt on the card

A bulk scan you approve in Agent Mode now sends every document the prompt on the card. When the prompt on a Bulk scan proposal card differed from the one in your Prompt Lab box, a live bulk scan (one not run as a background processing task) sent each document the prompt the box held before you selected Approve and start, while the task, Prompt History and the Prompt Log field all showed the card's prompt. A card started by the countdown in Give permissions did the same. Background processing tasks were not affected. Answers written before this release are not changed. See Agent Mode.

A cancelled background task no longer offers Resume Task

A cancelled background task no longer shows Resume Task. The button appeared on background tasks you had cancelled, but a cancelled background task is final (the tokens reserved for its unfinished documents are already back in your balance), so clicking it only ever returned an error. To scan the documents it did not finish, start a new bulk scan. See If a scan stops early.

Insight titles are always titles

A new Insight no longer takes a question as its title. Now and then, the automatic title of a new Insight was a question back to you (asking what you meant) instead of a title, and downloads of that Insight were named after it. Claira now checks each automatic title and tries again when what comes back is not a title; if it still is not, the Insight keeps Generating title… so you can regenerate the title or rename it. Titles already saved are not changed. See Insights.

Choose when Claira writes the prompt to the Prompt Log

Writing the prompt to each document's Prompt Log is now your choice on every bulk scan. When your case has a Prompt Log field, the bulk scan confirmation dialog shows a Write the prompt to each document's Prompt Log checkbox with that field's name. It starts unticked in cases you were already using, a case you open for the first time starts on your latest choice from your other cases, and scans the Agent starts for you follow it too. Until now, every bulk scan wrote the prompt whenever a Prompt Log field was set up. See Bulk Scan and Setting up the Log Field and the Prompt Log field.
Saving a single result gives you the same choice. In Prompt Lab and Multi-Code, a small Prompt Log section below the response (closed by default, its header showing On or Off) holds an Also write the prompt to the Prompt Log checkbox; the dialog that saves an Agent answer to a field has it too. The three share one choice, separate from the bulk one, with the same starting rule: unticked in cases you were already using, your latest choice in a case you open for the first time. None of these checkboxes appear unless the case has a Prompt Log field. Until now, every single save wrote the Prompt Log whenever a Prompt Log field was set up. See Recording the prompt in the Prompt Log and Agent Mode.
Background scans now write the Prompt Log too. With the box ticked, a background processing task adds the same Prompt Log entry a live scan does to every document it codes, and a live scan you send to the background keeps writing it without adding a second entry to documents already coded. Previously, background tasks never wrote the Prompt Log. See What gets written back.
Task details only name a Prompt Log field the scan actually wrote to. A scan started with the box unticked no longer lists the Prompt Log field in its task details.

Help articles and Agent Mode brought up to date

What Claira can do now describes Insight templates as they appear today: behind the Quickstart button on the Insights page, each with a Ready, Partly ready or Not run yet chip. It also describes Field Migration's four guided steps, instead of a Check selection button that no longer exists.
The French help articles on single review and on plans and billing now name the scan button Analyser le document, as the app does.
Asked where Claira runs, Agent Mode now says that your data stays in your deployment's region and that Claira only uses AI providers that meet its AI provider standards, instead of naming a hosting company. For the providers that serve your region, contact support@claira.to.

Admin Portal activity opens on what your team finished

The Activity screen now opens on Key activity. The switch above the type chips has two settings: Key activity, the work worth a look (a bulk task that finished, an Insight generated, a case assessment run, a report downloaded, a case activated or deactivated, a Case Context created or approved, an overage incurred, tokens bought, a plan changed, an invoice issued, a decision taken in the portal, a role assigned, an email that failed), and Everything, which adds the register underneath. Until now the feed opened on every document scanned and every prompt saved, and the Insight you were looking for was somewhere under them. See Key activity, or everything.
Nothing is ever dropped. When the quieter view is holding events back, it says how many under the chips, with Show everything beside the count. The download always saves exactly what you are looking at.
Every type chip now starts on, including Case Context and Email, which used to start switched off. Hiding a whole category was the only tool the screen had against volume; the switch does that job properly.
This release's features are in the feed too. Early case assessments have their own type. So do Report downloads: every Insight, Case Context, Bulk Task History report and prompt history someone downloads, with the format (PDF, Word or Excel), which answers "what left this case, and when?" without keeping a second copy of anything. The monthly invoice Claira issues appears under Billing and plans, and a change to a chat's permissions under Agent Mode. See Activity Feed in the Admin Portal.

Prompt PDF export: easier to read

The full version history PDF now gives the settled wording (the final version, or the current one when nothing is marked) the first page to itself, and starts the version list on the next page. Prompt text is set larger, and versions are further apart, so the document reads as a set of separate revisions rather than one column of text. Every section now opens with the same green-ruled heading the Case Context and Insight exports use. See Prompt History.
Both prompt PDFs are now titled Prompt export. The title used to be the first few words of the prompt, set large at the top of the page and repeated on every page after it, which said less than it looked like it did: the prompt itself is printed in full just below. The opening words still name the downloaded file.

Case Context PDF: the source documents heading is spaced like the others

In the Case Context PDF, the "Source documents" heading now uses the same spacing as the Insight and prompt exports. Where the section starts a new page, the heading keeps its space below the running title at the top of the page instead of sitting almost on top of it. Nothing else about the export changed.

Agent Mode: a file you attach is what your question is about

Attach a file to a question in Agent Mode and Claira now treats it as the subject of that question: it reads the file before answering instead of replying in general terms or asking you to paste its contents. Files from earlier questions stay available and are still used when your question touches on them. See Attach documents to a chat.

Admin Portal: what the open invoice figure includes

The Open invoice so far figure on the Admin Portal Overview now says what it is made of: hover the label to see the overage and the partner discount already included in it. The figure itself is unchanged, and it is the same total the Billing screen shows for that window: plan fees, overage and token packs, less your partner discount, before tax.

Use with context on Multi-Code templates

Multi-Code templates can now be used with your Case Context. Every Multi-Code template in Quickstart has Use with context beside Load into Multi-Code. Claira fills in the shared instruction and every field's prompt together, from your approved Case Context, and loads the whole template into the Multi-Code card with each field filled in. Field names, answer types and destination matching are unchanged, and your matter's facts go in the shared instruction rather than being repeated in each field. Most useful for relevance and privilege, where the template cannot know your issues, your counsel or your governing law. See Prompt templates.

Insight exports: the prompt moved to its own closing section

A downloaded Insight now opens on the Insight itself, and closes with what was asked. The PDF and the Word document used to print the prompt in a panel between the header and the text. That panel is gone: the report now runs header, Insight, Source documents, and then a Prompt section on a page of its own at the very end. It holds the prompt the Insight was generated from, headed Original, and, when you are exporting a later revision, each revision you asked for after it, headed Revision 2, Revision 3 and so on, oldest first. A first revision's export is headed Prompt and holds that one prompt; with revisions under it the section reads Prompt and revisions. See Downloading the Insight. The Excel workbook is unchanged: its Summary sheet still carries the prompt behind the revision.

Case Context: clearer selection limit

In Case Context, when you select more documents than the context still has room for, the header now says how many more you can add ("3 documents selected, you can only add 2 more") instead of naming a limit. The selection count is hidden on an approved context, which takes no more documents. See Setting up a case context.

Insight templates: no status on the tiles

The Insights Quickstart shows templates, not progress. The Ready, Partly ready and Not run yet chips are gone from the template tiles, and the details pane no longer opens with a line about which bulk run this case already has. A template is a starting point you may use once, twenty times or never, so Quickstart reads the same every time you open it: what each template produces, the prompt it reads, the prompt it puts in the box, and both ways to get its data (Run this prompt or Pick a field instead). Choosing a template still selects a matching completed bulk task as your source when the case has one, and the New insight card still tells you when the source you picked may hold nothing for it. See Insight templates.

Agent Mode sends the open document whole

The document you have open now goes to Agent Mode in full. While its chip is in the message box, Claira receives its extracted text and its own file (the PDF, image, audio or video) with every question. Writing "this document" no longer changes anything: the chip is the decision, and it is all of the document or none of it. See The document you have open.
The chip says so up front. It reads Text and file from the moment a document opens, instead of switching while you type, and narrows only where there is less to send: File for a document with no extracted text, Text in Local Mode, where the file is never sent.

Answer with Agent (Case Context missing information)

The Missing information list on a draft Case Context now has an Answer with Agent button beside Hide for now. It opens the context's Agent chat with the questions laid out and an answer box under each one: write what you know, leave anything you cannot answer blank, and click Send answers. The Agent places each answer in the section its question names and saves nothing until you confirm; questions you left blank stay on the list. See Answering the missing information questions.

Prompt save status points at Save Prompt

When the prompt cannot save on its own, Claira now tells you what to do instead. The row above the prompt box in Prompt Lab and Bulk Scanning used to offer Try again, which repeated exactly the request that had just been refused and so could go on failing indefinitely. It now reads Click Save Prompt. and points at the button beside it, which saves the prompt a different way. Your text stays on screen either way, and Save Prompt still adds the prompt to your prompt history.

Case Context action bar

Revise with Agent and Generate search terms moved out of the Case Context header and into the bar at the bottom of the context, to the left of Download Case Context, so every action on the context is in one place. See Setting up a Case Context and Generating Search Term Families.

Field picker: one marker for the field you chose

The field picker no longer marks your chosen field twice. In the Prompt Lab, Bulk Scanning, Multi-Code rows and the Agent's scan cards, the field you have selected carried both a check at the left of its row and a star after its name, for the same fact. The star is gone; the check stays, and in the Prompt Lab and Bulk Scanning pickers hovering it still tells you this is the field Claira will offer for your next scan of that kind. The used, not used yet and waiting-on-access markers are unchanged. See Bulk scanning.

Save status: the reason a change was not saved

When an automatic save fails, the status under your prompt or your Multi-Code setup now says which kind of failure it was. A Claira service that is momentarily unavailable reads as "Claira could not be reached", the same wording the panel already used when your settings failed to load; before, it read as "Claira did not accept the change", which suggested your text had been rejected when nothing was wrong with it. Your text is kept on screen either way, and Try again still sends it.
One failure now tells you how to clear it. If your settings for a case were created by a different Claira deployment, the automatic save cannot take them over and retrying is refused the same way each time. The status now says so and points you at Save Prompt in the Prompt Lab, which takes the settings over once and lets Claira save on its own again.

Insight progress shows three steps

The Insight window now shows three steps while an Insight is written, not four. The Checking Document IDs step is gone. It could only run for cases with background processing set up, so for everyone else it permanently read Document IDs not checked, which looked like the references had gone unchecked. They had not: every reference in a finished Insight is checked against the documents the Insight read, on every Insight, and a reference to anything else is removed before it is saved. That is the Checking references step, and it is unchanged. See Watching an Insight being written.

Insight revisions say they are rereading

When you revise an Insight, the first progress step now reads Rereading the original results instead of Reading the AI results. A revision works from the same values the first version was generated from, so your fields are not read again, and the step says so. See Watching an Insight being written.

Case Contexts attached to a chat

Agent Mode chats now show the Case Context they were given, beside the question it was sent with and in the chat's document list, with the Case Context icon. Click it to open that context. See Attach documents to a chat.
Generate search terms on a Case Context now attaches that context to the question it sends, so the Agent works from the context you started from even when your case has several, and the document you have open in Discover is left out of that question. See Generating search term families.
A chat opened with Revise with Agent shows the same chip on every question, naming the context it is revising.

Agent Mode header

Agent Mode says which mode you are in. The bar across the top of the chat window now carries the mode title, Agent Mode, with the same green tile and help icon as every other mode. It no longer reads "New chat", and it is no longer blank before you open a chat: once the chat has a title of its own, that title takes the place of the mode name, and you can still click it to rename a chat you own. See Agent Mode.
The chat window starts where every other mode's does. The gap between the top toolbar and the chat window was larger than in Prompt Lab, Bulk Scanning and the rest; every mode now begins at the same height, so switching modes no longer shifts the page.

Early case assessment: a clearer report

Early case assessment now fills in as it is written. The seven sections appear one by one while Claira writes them, each in its place, instead of arriving all at once when the assessment finishes. See Early case assessment.
The written assessment is now visibly separate from the counts behind it. What Claira wrote sits in a box of its own under The assessment; the statistics below it are now called Data analysis (they were "What it was built from"), on screen and in the PDF.
The fields an assessment counts moved into Settings. The standard pick lists are the same as ever and are named once in the article; the pick lists you add on top of them now live in a Settings section above the button, closed until you need it.
People - CC is counted instead of Document Created By, which recorded who processed a file rather than who wrote it.
Imports and ingestions are told apart. The headline figure reads Imports and ingestions, and every job now says whether it was an ingestion, an import from a load file, SQL or a database, or individual documents, with a tally by kind above the list.
The timeline has a marked axis in quarters, years or several years depending on how long the case runs, and the Spikes, Gaps and Busiest days are three labelled boxes under it.
Custodian coverage is now a bar per custodian, one block per year, darker where that year holds more of their documents and empty where they have none.
The largest files carry their Document IDs, on screen and in the PDF, so you can find them in Discover.
The PDF's Data analysis pages are now tables of one width, with bold, larger pick-list names and their values, counts and shares set out in columns.

The Agent and the early case assessment

The Agent knows whether this case has an early case assessment, and reads it when you ask about the case. Ask what the case holds, who the custodians are or where to start, and it answers from the latest assessment, naming the date it was run. Ask about an earlier one and it lists the assessments the case holds with their dates, so you can have it compare two runs and say what changed. See Early case assessment.
It can take you to the report. Ask to see an assessment and the Agent offers a button that opens it in the Early case assessment mode, on the run you asked about. In a case where none has been run, the button brings you to the screen with Run assessment ready for you, and Give me a first look at what's in this case joins the suggestions it opens with. The Agent still never runs one itself: the reading happens in your browser with that screen open. See Agent Mode.
While an assessment is being written, the Agent says so instead of describing a report it cannot see yet.

Insights: the revisions get out of the way

An Insight you have not revised now has the whole window. The list of revisions used to hold a column beside every Insight, even when the only thing in it was the original. It appears once there is a revision to choose between, and not before. See Insights.
Where there are revisions, they wait behind a tab. A small tab on the left edge of the viewer opens the list when you hover it, and steps back out of the way when you pick a revision, so the Insight keeps the full width the rest of the time. Click the tab to keep the list open while you compare revisions, and Hide revisions to put it away. The Revision dropdown that appeared on a narrow panel has gone with the column: the tab reads the same on a narrow panel and a wide one.

Bulk scan: the progress bar owns up to skipped documents

Documents set aside while Claira was loading the files now turn pink in the background progress bar, while the run is still going. A background Image, Audio, Video or Auto scan settles some documents before the scan itself starts, as it uploads each document's file. Those documents used to sit in the amber segment for the rest of the run, where the bar said they were prepared and waiting their turn. See Monitoring background tasks.
A document with no file the scan can read is now reported as skipped, not failed. In an Image, Audio or Video run, a document whose file is not a format that mode reads, or is larger than a scan accepts, is a skip, exactly as a Text run finding no readable text has always been: nothing went wrong, the document simply has nothing that run can look at. It no longer appears among the documents that could not be coded, and it is refunded as before. The note in the destination field still says which file was refused. See What counts as a successful document.
The amber label now follows what the run reads. It says File extracted on an Image, Audio, Video or Auto scan, where what Claira gathered is the document's own file, and keeps saying Text extracted on a Text scan. Where one legend covers runs of both kinds, it reads Ready to scan. The progress line reads Extracting files on those runs too.
Auto falls back to the document's text whenever the file turns out to be unusable, not only when it is too large. Auto chooses a file from the list Nuix Discover shows, so a file whose contents are not the format its extension promised, or that the chosen model does not read, is only discovered when Claira opens it. Auto now reads that document's text instead and charges it at the Text rate, in a background run and in your browser alike, rather than giving up on the document. See Media scans.

Agent Mode: ten files per chat

An Agent Mode chat now holds up to 10 attached files, where it held 5. Attaching several files to one question already worked, by dropping them together or picking more than one at the paperclip; what has changed is the ceiling. It is still a ceiling for the whole chat rather than for one message, because every file you attach stays with the chat and can be asked about turn after turn. Remove a file you are finished with to free a slot. Each file is still up to 10 MB. See Attach documents to a chat.

New pricing: more tokens on every plan, and a price per token that falls as you grow

Every plan now includes far more tokens. Pro goes from 6,250 to 25,000 tokens a month, Business from 71,500 to 300,000, and Enterprise from 500,000 to 2,000,000. Team is 85,000, and Starter keeps its 1,000. Prices are unchanged, so the price per token drops at every step: about 4.9¢ on Starter, 2.0¢ on Pro, 1.8¢ on Team, 1.7¢ on Business and 1.5¢ on Enterprise. See Plans and Billing.
Buying tokens is one price everywhere: $0.10 a token. Token packs cost the same on every plan, and the same as overage, so buying ahead and overrunning are priced identically and the plan itself is where the saving is. A 500-token pack is $50 for everyone. Because a plan's own tokens cost between 1.5¢ and 4.9¢, moving up a plan is the cheaper way to add real capacity.
Your plan's tokens are now a monthly allowance. On your renewal date the allowance is refilled to your plan's full amount, and what you did not use in the month just ended does not carry forward. Tokens you bought are not affected: packs, and any tokens refunded to you, sit in a separate part of your balance that never expires, and Claira always spends your plan's tokens first. The Usage tab shows both figures with the reset date. See Your plan's tokens reset each month.
The free daily tokens are now per person. Each user on a case can claim 100 free tokens once a day, rather than one claim for the whole case. Each plan sets how many claims a case can take in a day across everyone on it: 1 on Starter, 3 on Pro, 5 on Team, 10 on Business and 20 on Enterprise. See 100 free tokens a day, for every user.
Your new allowance arrived the day the pricing changed. Every active case was credited its new plan's full monthly tokens on the day the update shipped, whatever point of its billing cycle it was at, so nobody had to wait weeks for their renewal date to see the new amount. It appears in the Usage tab's activity list as New plan refill, and it sits on top of whatever the case was already holding.

AI provider standards: one declared exception for very long bulk prompts

Our AI provider standards now name one exception to "no caching", and it is ours to control. When a bulk scan uses a very long prompt (for example, a reference document pasted into it), Claira may keep that prompt with the AI provider for the length of the run instead of sending it again with every document. It stays in your deployment's country, is encrypted with Claira's own key, is deleted when the run ends, and expires on its own within an hour of its last use. The documents being scanned are never kept this way, and nothing changes in how you run a scan or in its results. See How We Choose AI Providers and Privacy and Security.

Bulk scan: scanning again with the same answer is no longer a failure

A document whose field already holds exactly the answer the AI just gave now counts as completed. Scanning the same documents again with the same prompt often gives most of them the same answer, and Nuix Discover has nothing to change when that answer is already in the field. Claira used to read that as a failed write and reported those documents with [E-WRITE]; it now checks the field and counts them as completed, charged like any completed document. The Prompt Log entry is checked the same way. Any other write that Nuix Discover does not confirm still fails with [E-WRITE]. See What counts as a successful document.

Saving a single result: no more "saved" when the field kept its old value

Apply in Prompt Lab, saving an Agent answer to a field, and Save to fields in Multi-Code now check that the value reached the field. Claira adds values and never overwrites them, so a field that holds one value and already has a different one is left unchanged. Claira used to report that save as done anyway; it now tells you the field already holds a different value, and you can clear the field in Nuix Discover and save again. In Multi-Code, the summary names the field rows left unchanged. Saving an answer the field already holds still counts as saved. See When the destination field already holds a value.

Claira Private: a case that runs entirely on your own model

Claira Private is a new way to activate a case. When you activate Claira, the standard plans are listed first: click the narrow Claira Private row beneath them, then Continue, and every AI feature in the case runs on your organization's own model, never on one of Claira's. See Claira Private and Case Activation.
Two ways to connect your model, both included. Set them up in Settings, under Your model: a Shared connection, an endpoint your organization runs that everyone on the case uses (anyone on the case can set it up), and a Local model, a model on your own computer that only you use. See Shared connection and Local model.
Every AI feature runs on it. Single reviews, bulk scans in your browser, Multi-Code, Agent Mode, Insights, the Prompt Generator, Revise with Agent and Case Context from documents. See What Claira Private includes.
One flat price, no tokens. Claira Private is $899 per case per month, with no tokens, no claims, no packs, no overage and no document limits. See Plans and Billing.
What works differently. Background processing is not available in Claira Private, scans are text only, and bulk tasks and chats get standard names.
Changing plans. A case cannot switch between a standard plan and Claira Private while it is active: deactivate it and activate it again with the other option.
Bring Your Own Model and Local Mode are now part of Claira Private. On the standard plans, the model picker no longer offers Bring Your Own Model and Local Mode is gone. See Model Selection.
New error messages explain each Claira Private situation, such as no model connected yet. See Claira Private errors.

Calmer live bulk scan pacing

Bulk scan: live bulk scans now run at a calmer pace at every Scan speed setting. Balanced, the default, works on about half as many documents at once as before, so results arrive in the live results panel at a pace you can follow and resource-availability errors are rarer. If you use your own model, the slider is unchanged: Fastest still uses the full number of concurrent requests you configured. See Adjusting the scan speed.

Case Context sections show formatted

Case Context sections now show formatted, as Insights do: labels and key facts in bold and lists as bullets, instead of plain text. Approved contexts show at full strength instead of greyed out. See Formatting in a section.
Click a section's text to edit it. It turns into a plain-text box with the cursor where you clicked, and a reminder underneath says how to bold words (**like this**) and make a bullet (a line starting with - ). Click outside it, or press Escape, to see it formatted again.
Claira and the Agent write sections in this format, and the PDF and Word downloads print the same bold and bullets. Bullets in existing contexts show as bullets straight away; bold labels arrive the next time Claira or the Agent updates the context.

Claira Private: a setup guide

A step-by-step guide for a new Claira Private case. Set up Claira Private takes you from activation to a first scan on your own model: choosing between a shared connection and a local model, setting either one up with the labels you see on screen, checking that it works, switching later, and where to look if something goes wrong. The Claira Private article, Case Activation and Plans and Billing link to it.

Revise with Agent and Answer with Agent leave the open document out

A chat opened with Revise with Agent or Answer with Agent no longer sends the document you have open in Discover. The chat is about your Case Context, so your answers go without it, and the document's chip in the message box starts out marked Not sent. Click Send again on it if you want the Agent to read that document too. Other chats send the open document as before. See Keep the open document out of your messages.

Claira Private: easier model setup

A shared connection sets up in one test. As soon as you enter the endpoint URL and API key, Claira lists the models your endpoint offers and chooses the only one when there is just one, so a single Test connection runs every check, calibration included, and Apply recommended appears straight away. See Set up a shared connection.
The shared connection now lives under Team settings, in its own Shared connection section, because it applies to the whole case. Your model, under Your settings, keeps your choice between it and a local model, and marks the one you use In use only when it can run scans, or Not set up when there is nothing to run on yet. See Shared connection.
Local model setup is clearer. The setting each local server needs (such as OLLAMA_ORIGINS for Ollama or Enable CORS in LM Studio) is shown before you detect, with Claira's address filled in; Use this server picks a detected server; http://localhost:11434 and http://localhost:11434/v1 are treated as the same address; and with the bridge window you confirm your model once, in the bridge, where it is already filled in. See Connect it in Claira.

Case Context keeps what you typed

Add to context no longer drops text you typed into a section. Over long documents, above all ones sent with their native files, Claira could rebuild a section from the documents and leave out what your team had written in it, such as the parties or counsel. It now keeps everything the context already says. When a document disagrees with it, for example it names a different respondent or seems to concern another matter, Claira keeps both versions, notes on the line that they differ, and adds a question to Missing information asking which one is right. See Generating content from documents.

Release 056 -- September 9, 2026

The Admin Portal opens as one console for your whole organization, Bring Your Own Model and Local Mode let a case run on your own endpoint or on a model on your own computer, Claira and its help centre now speak German, Insight templates and multiple Case Contexts shape the work around each matter, Prompt History groups a prompt's versions and exports them as PDF, Field Migration's field lists open quickly again, sovereign SCX.ai models arrive for Australia and Germany, and Gemini Flash 3.5 becomes the default model for Canada tenants, alongside a long run of corrections that makes usage totals, Insight citations, error messages and translations say what actually happened.

August 18, 2026 release video

Added the August 18, 2026 update video to Release Videos: a single-feature walkthrough of Smart Revise, the wand that turns a comment about an AI response into ready-to-apply prompt edits.
Rate responses and Smart Revise now opens with a video walking through the feedback buttons and a full Smart Revise pass, so you can watch it once instead of reading the steps.

PDF Unitization templates and workflow

Two new PDF Unitization templates in the objective coding section of the template library: PDF Unitization for compiled files whose document boundaries are self-evident, and PDF Unitization (Custom Split) for when you describe the split: by individual, record type, chapter, or email versus attachment. Both output the page-range string the Nuix Discover image viewer uses to split the file apart.
A new PDF Unitization workflow walks the full pass: have Claira draw the boundaries, QC and apply the split in the image viewer, then objectively code the new sub-documents (which inherit no metadata from the source file).

Field-creation rate limits removed

Removed the hourly and daily field-creation limits: the "Field-creation limit reached" message is gone, and you can create as many Discover fields as your work needs.

Connected fields listed alphabetically

Connected fields now appear in A-to-Z name order everywhere they are listed: the Connected Fields section of Settings and every field-selection dropdown (single review, bulk scan, Agent Mode and the save-to-field dialogs), instead of the order the fields were connected in, so a field is always where you expect it in the list.

Insight PDF source-documents section

The Source documents section of the downloaded Insight PDF now opens with a prominent heading and a one-line note explaining that the listed documents (and the AI response shown under each) are what Claira analyzed to generate the Insight, so readers of the export immediately understand what the trailing pages contain.

Prompt History version grouping

Prompt History now opens in a Grouped view that stacks variations of the same prompt into version chains: expand a card to read the progression oldest-to-newest, with added words highlighted, removed words struck through, and a +N −M words chip per step. Identical re-runs collapse into one entry with a Used N× count, and a toggle switches back to the classic chronological list (your choice is remembered per case).
You can now mark a version as final for its chain: a shared, case-wide green Final badge that makes that version the group's representative and what its Restore button inserts. Anyone on the case can set or move the mark, and the prompt's details show who marked it and when.
Your Grouped / Chronological choice is now remembered even before you have written a prompt. It no longer waits for a prompt to be saved with your settings, so it follows you to your other devices from the first time you flip it.

Insert into AI Review jumps to the Prompt Lab

Clicking Insert into AI Review on a prompt the assistant formats in Agent Mode now takes you straight to the Prompt Lab with the prompt already in place, and the prompt box flashes green for a moment so you can see exactly where it landed. No more silent insert you had to go looking for.

Gemini Flash 3.5 is the new default model for Canada tenants

Canada tenants now scan on Gemini Flash 3.5, the newest Gemini Flash generation, hosted inside the Canadian data residency boundary. It replaces Gemini Flash 2.5 as the default in the AI Settings model picker and brings a much larger context window, so long documents that previously had to be truncated now fit in a single pass. See Model selection: Available models.
Gemini Flash 2.5 stays available in the picker, now labelled with its version number so the two Flash generations are easy to tell apart. Cases that were already set to a specific model keep that model: nothing is switched mid-review, and your plan is charged the same per document whichever model you choose.
Australia tenants are unchanged: Gemini Flash 3.5 is not offered in the Australian region, and Gemini Flash 2.5 remains the default there.

Multiple Case Contexts with templates

Case Context now supports multiple contexts per case: one per workstream (up to 10). The Case Context panel opens on a list of the case's contexts; each keeps its own name, sections, source documents, and Draft/Approved state, and one is marked with a star as the case default.
New contexts are created from templates that match your world: Litigation / eDiscovery (a Case Context), Regulatory investigation (a Review Mandate), Internal investigation (an Investigation Brief), Law enforcement (an Operation Brief), or Start blank (a Review Context). Every template starts with three or four sections, and sections can be renamed, removed, or added (up to 12 per context), so regulators, investigators, and other non-litigation teams get terminology that fits.
The Prompt Generator's "Include Case Context" checkbox is now a context picker: it lists the case's approved contexts, preselected to the default, with a "None: generate without context" option. See Setting Up Case Context.

Link previews for every help article

Sharing a help article (in Teams, Slack, an email, or on social) now shows a preview card with the article's own title, its summary and the section it belongs to, instead of the same Claira logo on every link. Nothing to do on your side; existing links pick the new card up automatically.

Cleaner loading throughout the app

Replaced spinners and "Loading…" text with skeleton previews throughout the app: the conversation list and chat, bulk scan task history and task details, past insights and the insight window, Field Migration, usage statistics, settings, plan choices, and the template, prompt-history and comment windows now sketch their final layout while content loads, then fill in without anything jumping.
Refreshing a list no longer blanks it: what's on screen stays put while the refresh runs, and loading previews only appear the first time a panel loads. For users with reduced motion enabled, the previews hold still instead of shimmering.

Multi-Code results clear when you open another document

Fixed Multi-Code single-document results staying in the panel after you opened a different document in Nuix Discover. The response, the extracted field values and the feedback buttons now clear as soon as the new document loads, the same way the Prompt Lab response already did, so Save to fields can no longer write the previous document's values onto the document you just opened. A Multi-Code run still in progress when you switch documents is cancelled instead of filling the panel back in.

Case Context PDF source-documents section

The Source documents section of the Case Context PDF export now opens with a prominent heading and a one-line note explaining that the listed documents are the ones Claira read to draft the context, and each document is shown as its title with the document ID beneath, matching the Insight PDF.

Field Migration

The From and To field lists in Field Migration now open much faster on large cases. Loading the lists no longer asks Nuix for a per-field document count that was never shown on screen.
If the field lists ever do fail to load, Field Migration now says so and offers a Refresh button instead of showing a spinner indefinitely.

Admin Portal

A new Admin Portal at app.claira.to gives organization admins one console for everything that spans your whole organization. Open it from the Claira menu in Discover (Manage organization), with no separate sign-in. See the Admin Portal overview.
Cases: every case Claira has been activated in, with plan, reservoir, usage and the lifecycle dates you reconcile an invoice against, plus a detail page per case.
Users and Roles: who uses Claira, one role per person, roles you define with permissions picked from the full list, an access-review export, and enforcement you switch on when you are ready.
Billing and invoices: the partner invoice per window with grouped lines, overage and the partner discount, a CSV export, the cards Stripe holds, pending plan changes, invoice parameters (discount and window day), and a Billable switch to keep test cases off the invoice.
Email routing: decide which roles receive each kind of email for all your cases at once, and read a delivery log that also lists suppressed sends.
Activity: one feed of everything that happened across your cases and your team, with type chips, search, filters and an export.
Settings and Discover portals: rename your organization, connect another Discover portal yourself and replace its auth token.
Users and Permissions now explains the split between Discover's case access and Claira's roles.
The Roles screen now says which permissions Claira actually checks today. Permissions no action checks yet are marked Not yet enforced (you can still record them on a role, and they take effect as their checks ship), and the enforcement card states how many of the settable permissions are checked, so switching enforcement on means exactly what it says. Counts now say "of 129", the number of permissions a role can be granted. See Users and Roles.
Last active on the Users screen now means the last time the person opened Claira or ran something in it (a scan, bulk task, prompt, Insight, chat or field migration), not only a portal visit, so an active reviewer no longer reads "Never". Hover the column header for the definition; a person's details also show their last Claira activity on its own.
When a change does not save, the message now says what did not change and what to do next (for example, "That role change didn't save. Priya still has the role they had before. Try again.") instead of a bare "was not saved". A failed page load shows a plain sentence for your situation rather than an internal error code.
The portal is friendlier to screen readers: every loading screen announces what it is loading, confirmations and errors are announced when they appear, the Cases and Billing tabs work with the arrow keys, and the Roles areas announce whether they are expanded.
Excluding a case from invoicing now tells you when the case has no billing record yet (it joins the invoice once a plan is set on it in Discover) instead of reporting a change that did not happen.

Bring Your Own Model

Connect your organization's own model to a case: the new Bring Your Own Model section in Settings takes an OpenAI-compatible endpoint URL, a model ID, and an API key, tests the connection with named checks (URL, reachability, model discovery, completion, streaming, long context, Multi-Code, and a calibration run), and saves it once the essentials pass. The connected model then appears in the model picker under the display name you chose, and the case's single reviews, live bulk scans, and Multi-Code passes are sent to your endpoint. See Bring Your Own Model.
Throughput is tuned per connection: Concurrent requests starts at 1, the calibration check measures how many parallel requests your endpoint sustains with stable latency and offers an Apply recommended button, and the scan speed slider scales to that capacity during live bulk runs. Claira eases off automatically when your endpoint starts returning errors and flags the connection as Needs attention. See Optimizing your connection.
Endpoint problems get their own messages: "endpoint not responding" (resume the task once the endpoint is back) and "API key rejected" (Claira stops sending immediately; replace the key with Test & save), plus a Diagnostics view of recent endpoint errors in Settings. New entries under Model endpoint errors explain each one.
In this release your own model handles text scans in live runs only: media scans and background processing stay on Claira's models, as do Agent Mode, Insights, Case Context, Smart Revise, the Prompt Generator, and automatic titles. The endpoint must be publicly reachable over HTTPS with a publicly trusted certificate. Per-document billing is unchanged. Model Selection and Privacy and Security describe how a connected model fits in.

Local Mode: run Claira on a model on your own computer

Local Mode is a second way to bring your own model, and this one never sends your documents anywhere. Turn it on in Settings and Claira runs every review on a model on your own computer (Ollama, LM Studio, vLLM, or llama.cpp), with your browser talking to it directly. While it is on, Claira's servers make no model call of any kind for you on that case. See Local Mode.
Setup is a detect-and-confirm flow. Claira looks for a model server on the usual ports, lists the models it offers, runs a short set of checks (can it complete a review prompt, return well-formed JSON for Multi-Code, use function calling, stream, and how much can it handle at once), and shows you a PASS, WARN, or FAIL row for each. Your endpoint address and its optional key are saved in your browser only; they never reach Claira's servers. Supported local servers lists the one CORS setting each server needs.
Your browser asks permission once. Chrome now prompts before a website may reach your own network. Choose Allow. Administrators can pre-grant it for everyone with the LocalNetworkAccessAllowedForUrls policy, pointed at the Claira address. See The browser permission.
A small bridge window does the talking. Because Claira runs inside Nuix Discover as an embedded panel, browsers will not let the panel itself reach your network, so Claira opens a separate window at its own address for that. Keep it open while Claira is working: it shows the endpoint, your model, the permission state, and a live request count, so you can see your own model doing the work. See The bridge window.
Nearly everything works. Single reviews stream as usual, live bulk scans keep the worker pool, speed slider, live feed and resume, and Multi-Code runs when your model passed the check. Agent Mode, Insights, Case Context from documents, the Prompt Generator, and Smart Revise all run on your local model too: Claira prepares the request, your model answers it, and Claira reads the answer back. Only two things are switched off: background scans and image, audio, video and PDF scans. Both controls appear disabled with a short note and a Why? link. See What works in Local Mode.
You always know you are in it, and leaving is deliberate. A Local model badge sits in the app chrome with a status dot, the model picker shows your local model as a locked value, and Leave Local Mode asks you to choose a Claira-hosted model before it will switch back. Local Mode is yours and per case: a colleague in the same case is unaffected. See Leaving Local Mode.
Document content never leaves your machine for inference. Claira still stores your prompts, results, and the usual metadata. Privacy and Security is precise about the difference. New messages for a closed bridge window, an unreachable model, a blocked browser permission, and a document too large for your model are documented under Local Mode errors, and Troubleshooting Local Mode covers the rest.
Billing is unchanged: documents scanned on your local model cost the same per-document tokens as documents scanned on Claira's models, and single reviews stay free.
Local Mode never fails open. If Claira cannot confirm whether Local Mode is on for your case (a momentary database hiccup, say), it now refuses the request and tells you that nothing was sent to a cloud model, instead of quietly treating Local Mode as off for a minute. The same rule covers Insight title regeneration, which is now blocked in Local Mode like every other cloud feature. See Local Mode errors.
Background scans pause while Local Mode is on. A background scan that was already running when you switched Local Mode on now pauses, so no further documents are sent to Claira's models, and resumes where it left off when you switch Local Mode off. See What works in Local Mode.

Connected model improvements

Azure OpenAI endpoints can now be connected. An endpoint URL may carry the ?api-version= parameter Azure requires, and Claira sends it with every request. See Endpoint URL.
The Multi-Code check no longer gets skipped. It now runs before the optional checks, so a slow endpoint can no longer leave Multi-Code switched off without telling you why. See Testing the connection.
Connection tests are limited to 10 per hour for a case. A saved connection keeps working normally; only running the test again is refused past that.

Insight templates

Insight templates. Five common case outputs (Chronology, Early case assessment, Key documents, Issue analysis and Privilege log) now appear as a row of templates at the top of the New insight card. Picking one fills in the source and the prompt, both still fully editable. Writing your own prompt works exactly as before. See Insight templates.
A finished bulk task offers the insight it can produce. When a bulk task that started from a prompt template completes, it offers the matching insight directly: select Generate insight and the compose card opens already filled in. Not now removes the offer for you permanently, and it never reappears once the insight has been generated.
Templates tell you what they need. A green dot marks a template whose data your case already has. Without one, Claira opens Bulk Tasks with the suggested prompt filled in so you can choose the documents and confirm the cost there. Claira matches on the template a task was started from, so editing the prompt wording does not break the link.

Browser and workstation requirements

A new Browser and Workstation Requirements article gives your IT or desktop engineering team everything they need in one place: the supported browsers (current stable Chrome or Edge), the single outbound rule Claira needs (HTTPS on port 443 to *.claira.to), and a note that there is nothing to install on a workstation.
The article calls out early that Nuix-hosted Discover environments need nothing special: the on-premises section is only for teams running Discover on their own infrastructure.
For on-premises Discover, it documents the two things that must be true before document text will load: the Portal-API has to allow the Claira origin on both the preflight and the request, and the browser has to allow the call into your internal network. It explains why "it works in Firefox but fails in Chrome and Edge" points at the second one, and walks the fixes in order: an internal hostname for Claira (recommended, and arranged with support), asking Nuix to delegate the permission on the extension frame, and browser policy as a last resort, including the common mistake of allowlisting the Claira address instead of the Discover one.

German in the Admin Portal

The Admin Portal now speaks German, alongside English and French. It follows your browser language, so a German-language browser lands on a German portal: every screen, every column, every tooltip. Language stays independent of region: a German interface over Canadian-resident data is a normal setup, and prices remain in your region's currency.
The German versions of the Admin Portal articles now quote German screen and button labels. They previously spelled every label in English, because there was no German interface to point at.

Tokens used is now net of refunds, and overage reads as a charge

The monthly tokens-used figure now subtracts refunds. When a bulk task is refunded (documents that failed, or a task you cancelled before it finished), those tokens came back to your balance but were still counted in Usage as if you had spent them, so the month's total only ever grew. It now reflects what you actually kept, and the per-user chart nets the same way, so the two agree. A refund counts against the month the original task was charged in, even when the refund itself lands later. Expect totals for recent months to be lower than before: the tokens were never actually billed, only miscounted on the screen.
Overage no longer looks like a credit in the billing ledger. An overage row was shown in green with a +, exactly like a refund or a token purchase, even though it records going past your reservoir rather than adding anything to it. It now appears in red with no sign, so a charge can no longer be mistaken for money back. See Usage Analytics.

German language support

Claira now speaks German. Choose Deutsch from the language dropdown in Settings, and the whole interface, the AI's answers, and your notification emails follow. The picker replaces the old English/French pair of buttons, and your choice follows you across browsers and devices. Scans, Insights, and Agent Mode replies come back in the language you have selected.
German is available to every customer, in every region. Your language and where your data lives are separate choices: data residency continues to follow the Claira address you open, and nothing about it changes when you switch language.
Dates, numbers, and amounts now format to the language you picked, so a German interface shows German date and number conventions whichever currency your plan bills in.
The whole help centre is available in German, every article, on German URLs. The language switcher on any article takes you to the equivalent page in the other languages. The Quickstart prompt-template library is German too: names, descriptions, and the prompt text itself.
Your language choice now follows you across browsers and devices, as promised. It is saved the moment you pick a language, and Claira opens in that language wherever you next use it. Previously the choice stayed in the browser you made it in, so a new browser or device came back in English.
German everywhere the interface reaches. The connected-model panel under Settings, the prompt name and prompt text behind an Insight template's Run this prompt, and the in-app help links (Open docs on an Insight template, the model-settings tooltip, the plans and setup links on the activation screen, and the help links in your notification emails) now follow a German interface instead of falling back to English.

Sovereign models from SCX.ai (Australia and Germany)

Six new models, hosted by SCX.ai. Australia tenants gain MAGPiE (Australian sovereign), Llama 4 Maverick, GPT-OSS 120B (SCX) and Gemma 4 31B; Germany tenants gain Qwen 3.8 Max and GLM 5.2. SCX.ai is an Australian company running its own hardware in its own data centres, so for Australian tenants both the data and the operator stay onshore, the distinction that matters when a procurement requirement asks for an Australian-owned provider rather than an Australian region of a global one. Pick any of them in Model Selection.
Germany gets a much larger context window. Qwen 3.8 Max and GLM 5.2 both accept far longer documents than the Gemini Flash models, so documents that previously had to be split can be scanned whole.
Every model in the picker now shows its country flag. Whichever model you choose, your documents are processed inside your own data residency boundary. The flag lets you confirm that at a glance without leaving Settings.
Switching to any of the new models is a model-name change only: your prompt, fields, and case settings carry over, and your plan is charged the same per document.

Why documents fail during a bulk scan

The failed counter on a live bulk scan is now hoverable. Point at it and Claira explains, in plain language, that a few failures are normal on a large run, what usually causes them (a media file that could not be retrieved from or written back to Nuix Discover, a document the AI could not process because it was too long or was blocked by a safety filter, or the AI service being under heavy demand), and that failed documents are never charged: their tokens are refunded automatically when the task completes. See Bulk scanning.

Download a prompt as PDF

Any prompt can now be downloaded as a PDF. A Download PDF button appears on group cards and history rows, and in a prompt's details: the document to attach when you send a prompt out for approval, or to file the wording that was actually run. See Prompt History: Downloading a prompt as PDF.
Prompts with more than one version offer a choice: This version only, a short document with the prompt's text, its author and dates, its place in the chain and its comments; or Full version history, which walks the whole chain oldest-first, showing what changed at each step (added words in bold green, removed words struck through in red, with a +N −M words count), followed by that version's complete text.
The review conversation comes with it. The full-history PDF carries every comment left anywhere in the chain, grouped under the version it was left on, so a reader can see why the wording ended up where it did.
The export carries the Claira header and styling used by the Insight, Case Context and Bulk Task History PDFs, plus the case name, who exported it, and numbered pages.
Every export is recorded. Downloading a prompt writes an audit entry (who exported it, which prompt version, when, and from where), kept for your organization's compliance records. The entry is written before the file is handed to you, so if it cannot be recorded the download is not created and Claira tells you to try again. Nothing from the prompt or its comments is copied into the audit entry.
Exports are far smaller. The Claira logo in every PDF (prompt, Insight, Case Context, Billing Report and Bulk Task History) is now compressed and stored at the size it is printed, so a one-page export is a few tens of kilobytes instead of nearly 3 MB.
Characters the PDF font cannot show are marked, not dropped. Text outside the Western European character set (Chinese or Cyrillic names, mathematical symbols such as ≥) used to vanish from the PDF or garble the whole line, with no warning. Each such character now appears as ¤ in the document, the document's last line says how many were replaced, and Claira tells you after the download. Accented letters, the euro sign, œ and « » are unaffected. This applies to every Claira PDF; for a prompt export the audit entry records the count too, so the record never claims a faithful copy that was not made. See Prompt History: Downloading a prompt as PDF.
Case Context filenames are localized and name the context. A Case Context PDF is now saved as, for example, Claira-Case-Context-Review Mandate-DRAFT-20260908.pdf, in French Claira-Contexte-du-dossier-Mandat de révision-BROUILLON-20260908.pdf, in German Claira-Fallkontext-Prüfungsauftrag-ENTWURF-20260908.pdf, instead of the same English name, built from the case name rather than the context's, in every language.
An Insight exported with no source documents no longer ends with an empty Source documents page.

The prompt generator sticks to the review you asked for

A short request now produces a short prompt. Typing something like "relevance review" into the prompt generator used to come back covering several things at once (a relevance pass, a tag for every issue in your Case Context, often a privilege pass as well), and trimming it down took longer than starting over. The generator now builds the prompt around the review you named, and nothing else.
Your issues still do all their work. The issues in your Case Context define what relevant means for your matter, so they are written into the definition of relevant rather than turned into a second set of codes to apply.
Ask and you get it. Add "and tag which issue each document goes to" and the prompt includes issue tags. Ask for "relevance and privilege" and it covers both. Ask for a "full review of everything" and it stays wide. See Case Context troubleshooting: Generated prompt covers more than you asked for.
Summaries, translations and chronologies are unaffected: the narrowing only applies to coding reviews.

Generated prompts now describe each document, so Insights have something to read

Every prompt the generator drafts asks for a one-line description of the document. One or two sentences saying what the document is, who it is between, when, and what it concerns, on its own line, on every document, including the ones coded not relevant. See Prompting Overview: Ask for a short description of the document.
This is what makes an Insight worth running afterwards. An Insight can only work from what your prompt asked the AI to write down; a finished task full of bare Relevant / Not Relevant codes gives it nothing to synthesize. You no longer have to have decided you wanted an Insight back when you wrote the prompt.
Left out where it would do harm. When the whole answer has to be a single value going into one field (a date, a name, a document type, a translation), the description is omitted, because a second sentence there would corrupt the value.
Prompts you wrote yourself are untouched, and so are tasks that already ran.

Admin Portal: portal registration and token rotation

A Discover portal address can be registered once. Connecting a portal in Settings → Discover portals is refused when that address is already registered in Claira (by your organization or by anyone else), so no organization can attach its own auth token to a portal it does not hold. A replaced auth token now reaches every Claira server within a minute: sessions signed with the new token are accepted straight away, and the previous token stops being accepted everywhere within about a minute. See Settings and Discover portals.

Case deletion

A scheduled deletion now removes more of the case. Alongside the review history it always removed, running the deletion now also takes out any connected model (Bring Your Own Model) set up for the case together with its stored API key, the records of Local Mode runs, the pin, final-mark and export records of the case's prompts and the pin and sharing records of its chats, the field-creation records, each user's settings for the case, the case's email recipient list, and any Discover service credential saved for background processing, including the stored token. Billing history, the token ledger and the deactivation / deletion audit trail are still kept. See Deleting a case's data: What gets deleted.
Local Mode transcripts are cleared when a run finishes. A Local Mode run's conversation with your model (including any document text its tools read) is dropped the moment the run ends, whether it completed, was cancelled or timed out, and finished runs are removed after a short retention period. See Local Mode.

Field creation: no per-case cap, every creation audited

There is no cap on how many fields you can create in Discover from Claira: the "creation is paused while too many created fields await administrator access" safeguard described earlier never applied and has been removed from the app and the documentation. Every creation attempt and its outcome is recorded in the case's audit trail.

Insight citations are only shown as verified when every source document was confirmed

A citation in an Insight now renders in the verified (green) style only when the document it cites was confirmed against Nuix Discover while the Insight was generated. Previously, when Nuix Discover returned only some of the source documents during that check, the ones it did not return were still shown as verified, in the viewer and in the PDF's "Citations verified" count. They now render red with a strikethrough, like any other citation Claira could not vouch for, so you may see fewer verified citations on new Insights: that is the check being honest, not the Insight getting worse. Existing Insights are unchanged.

GPT-OSS 120B now works from the French interface (Canada tenants)

Canada tenants using Claira in French can now scan with GPT-OSS 120B. Choosing it in Model Selection while the interface was in French returned an Invalid model error on every scan; the model now runs from the French interface exactly as it does from the English one. The model, its Canadian residency and the per-document billing are unchanged.

Prompt History: removing a final mark works from the grouped view

In Prompt History, Remove final mark now works from the Grouped view even after that version has been re-run. Previously the mark could only be cleared from the classic list once an identical re-run had been saved, and the app reported success while the Final badge stayed. Opening a final version's details from the grouped view now shows it as final too.
If a teammate has already moved or removed the mark you are clearing, Claira tells you the mark had changed and refreshes the history instead of reporting a removal that did not happen.
For the organization's compliance records, every attempt to mark a version that is already final is now recorded alongside the standing mark, and removing a mark records whose mark was removed.

Insert into AI Review behaves like every other insert

Insert into AI Review in Agent Mode now goes through the same path as inserting a template or restoring from history: the prompt is saved to Prompt History right away, so it is there when you open the history from the Prompt Lab, and the next bulk task is no longer attributed to a template the prompt did not come from.
If the prompt box already holds a draft, Claira now asks whether to Replace or Append instead of overwriting your work.
Jumping to the Prompt Lab no longer changes which mode Claira opens in next time: that setting only changes when you switch modes yourself.

Clearer messages when an Insight cannot be created, deleted or loaded

If deleting an Insight fails, Claira now says so under the Insight's title instead of closing the confirmation silently and leaving you to wonder whether it worked.
Starting or revising an Insight when your case is out of tokens now says that the Insight would exceed your remaining balance, with a link to Plans & billing being the next step, not "please try again".
If the list of past Insights cannot be loaded, Claira now shows an error with a Try again button rather than telling you the case has no Insights yet.
When an organization uses Admin Portal roles to restrict who may delete Insights, disconnect fields, revoke a background-processing service account or schedule case-data deletion, a refused action now explains in your language that your role does not allow it, instead of showing an untranslated technical message or nothing at all.
A failed Prompt Generator run shows a plain "Failed to generate prompt" in your language instead of the AI provider's raw error text.
Opening an Insight that cannot be loaded now shows the reason with a Try again button instead of a loading placeholder that never finishes, and screen readers are told when the Insight and its revisions have finished loading.

Clearer failures, honest promises

A field list that failed to load now says so. If Claira cannot fetch the case's connected fields, every field picker (Settings > Connected Fields, the Prompt Lab and Bulk Tasks dropdowns, the Multi-Code destination picker, the Agent "save to field" dialog and the Insights field picker) shows "The case's fields could not be loaded" with a Refresh button, instead of an empty list that invited you to create a field the case may already have.
Case Context errors are plain sentences. A failed create, save, rename, default change, deletion, approval change or load now tells you what did not happen ("Your Case Context changes could not be saved…") rather than an internal code or a message borrowed from another feature.
The bulk confirm dialog only promises an email when one can be sent. When you choose "Run as background processing task", Claira checks for your email address as the dialog opens: it says "You'll get an email when it finishes" only when it found one, and tells you plainly when it did not. The "safe to close" panel does the same. If a background task finishes or fails and Claira had no address for the person who started it, the task's detail view now notes that no email was sent.
Bulk failure notes match your scan settings. The advice appended to a failed document's note in the response field now follows the task's Scan-as setting: a media scan that failed no longer tells you to enable a media scan. This changes the wording written into the field for failures from now on; existing notes are untouched.
Manage plan recovers from a failed load. Settings > Manage plan no longer shimmers forever when the plan catalog could not be fetched; it says the details are unavailable and offers Refresh.
Field Migration's Cancel button can be retried. If a cancel is refused because the migration already finished, or the request fails, the button re-enables and says which.
Accessibility: Claira's notification toast is now announced by screen readers (errors interrupt, other messages wait), its close button is named, and warnings use a triangle icon so they are distinguishable from errors without colour. The Info icons in the Local Mode and Bring Your Own Model settings are now reachable by keyboard and open a panel whose "Learn more" link can be tabbed to.

Local Mode: Confirm and start now checks your endpoint

The bridge window's Confirm and start now contacts your model before saving. If the server does not answer, refuses the request, or rejects your key, the window tells you why (the same message a scan would show, including the CORS hint when that is the cause), and keeps the previous endpoint (or none) until you fix it. The button shows a pending label while it waits. See The bridge window and Troubleshooting Local Mode.
"Last error" in the bridge window now records a refused confirmation, so the diagnostic the troubleshooting steps depend on is there the first time.
A browser that refuses to store the endpoint (a private window, or blocked site data) is now told so, instead of being asked to try again.

Connected model: health that survives a server restart, in plain words

Connection health no longer depends on which Claira server answers. A rejected key or a run of failures seen anywhere now shows on every refresh, with no more "No recent activity" on one reload and an error on the next, and it clears once your endpoint answers successfully again (a passing Test connection with your saved key counts). See Optimizing your connection.
The health line now says what went wrong ("Last problem: your endpoint asked Claira to slow down · date") instead of printing an error code. The code itself is still in Diagnostics, where it belongs.

French accents restored throughout the app

French text that had lost its accents now reads correctly: the Agent Mode chat copy (the welcome line, the bulk-scan status notes, the Insérer dans la revue IA pill), the activation and plan screens, the Multi-Code conversion messages, the "Généré par Claira" line in a downloaded Case Context PDF, and the messages shown when a Discover session cannot be verified.

Interface words that had stayed English

The Reasoning level picker in AI Settings now offers its levels in French and German (faible / moyen / élevé, niedrig / mittel / hoch) instead of the English words, the template library's Try again and No templates available states are translated, a German Insight PDF is subtitled Claira-Insight, and the plan manager's upgrade button reads Upgrade durchführen in German. The old activation-screen line quoting a fixed dollar price is gone. Every price you see comes from your plan in your region's currency.

Agent Mode approvals in every language

In Agent Mode, a short reply of oui or ja now approves a proposed bulk scan the way yes does, and the Insert into AI Review pill beside a scan proposal behaves the same in French and German as in English: it stays hidden while the proposal card offers its own approve button.

Help Center links stay in your language

Links inside the German Help Center now open the German article. Cross-references in the German help pages (the German changelog included) used to open the English version of the linked article. They now take you to the German page, and to the matching section when the link points at a heading.
Trimming a German help address back to its section no longer shows "page not found". …/ki-sichtung/prompting now opens the Prompting overview, as the English and French addresses already did; the Emails and Analytics section addresses open their first article in every language.

Release 055 -- August 28, 2026

Claude Sonnet 4.6 arrives for Australia tenants, busy-model errors get a one-click Change model button, and background processing becomes available under every model.

Claude Sonnet 4.6 for Australia tenants

Australia tenants can now select Claude Sonnet 4.6 in Settings, hosted on AWS in Sydney inside the Australian data residency boundary, a strong alternative when the Gemini models are experiencing high demand. Claude handles text scans (single, Multi-Code, and foreground bulk); media scans continue to require a Gemini model. See Model Selection.

Clearer messages when a model is busy

When a scan fails because the AI model is experiencing high demand, the error message now appears in your interface language and, when another model is available in your region, includes a Change model button that takes you straight to the model picker in Settings.
Scan failures that previously showed raw technical details in the notification now show a plain-language message instead.

Background processing under any model

The Run as background processing task option is now available regardless of the selected model. Background processing always runs on Google Gemini, so when a non-Google model is selected the confirmation dialog shows a note that the background run may use a different model than the one selected. See Model Selection: Models and Background Processing.

Release 054 -- August 11, 2026

Insight reports grow richer with per-document citation checking, Field Migration matches values into pick lists and splits memo fields, "Use Multi-Code" now writes real field prompts, new thumbs and Smart Revise let you rate and refine AI responses, and plan changes move entirely into the Settings plan manager.

Clearer background-scan progress bar

The color-coded progress bar for background bulk scans now uses plainer labels: green is Complete, blue is Scanning, amber is Text extracted, and pink is Skipped. The old Awaiting write-back (teal) band has been folded into Scanning, so documents finishing AI and waiting to be written back now stay in the blue segment instead of a separate color. The bar also sweeps in from left to right when a task's progress first loads. See Bulk scan: Monitoring background tasks.

July 21, 2026 release video

Added the July 21, 2026 update video to Release Videos: a short walkthrough of creating Discover fields from inside Claira and the new Author and Recipients coding templates.

Trustworthy Insight citations

Insight citations now verify each document ID individually: in a citation listing several documents, any ID that is not among the Insight's source documents is shown in red with a strikethrough while the verified IDs stay green. See Insights: Reading and revising.
The downloaded Insight PDF now carries the same warning signal as the viewer (unverified citations render red and struck-through instead of green), plus a new footer line, "Citations verified against the source corpus: n of m". See Insights: Downloading the PDF.
The Source documents section of the PDF and the Show sources panel now always pair each document's title, ID, and AI-response excerpt correctly, and the excerpts (previously blank for Insights built from a document selection) are included again. Document titles containing characters like & also display correctly.
Behind the scenes, Claira now rejects malformed or duplicate document IDs when a scan or an Insight is created, and, on cases with a saved Nuix Connect credential, re-checks document IDs against Nuix when a scan is created and again when an Insight is generated, automatically correcting any mislabeled citations it finds.

New AI Review article: variability in AI outputs

Added Why you might not get the same answer twice to AI Review: a plain-language explainer on why a large language model can word the same answer differently run to run, why vague prompts vary while structured relevance prompts converge, and why decisions should stay stable even when summaries don't.

Prompt Generator explains its disabled button

The Generate Prompt button in the Prompt Generator now explains itself when it is greyed out: clicking or hovering the disabled button shows a tooltip asking for a little more detail (descriptions need at least 10 characters). Previously the button simply did nothing, which could look like the generator was broken.

Background-scan legend explains itself

The color legend above background bulk-scan progress bars is now interactive: hover (or keyboard-focus) any legend entry (Complete, Scanning, Text extracted, or Skipped) for a short plain-language description of what that step means. The Complete segment also now uses Claira's signature green. See Bulk scan: Monitoring background tasks.

Richer, more detailed Insight reports

Insight reports now default to a fuller, more structured write-up: findings are developed in more detail instead of compressed into a brief abstract, longer reports are organised under section headings, and key names, dates, amounts, and findings are highlighted in bold, in the viewer and in the downloaded PDF alike. Your prompt still has the final say: ask for a specific format or length (or revise with "Make it shorter") and Claira follows it. See Insights.

Multi-Code handles empty values gracefully

When a Multi-Code field has no answer in a document (an email with no contract date, for example), Claira now shows it as a neutral Empty result instead of a technical "no value" warning, in the single-document response card and in the bulk-scan live feed. The Nuix Discover destination field is left unchanged, never overwritten with a blank. See Objective coding: Empty values.
A document where every Multi-Code field comes back empty now completes normally in bulk scans (foreground and background) instead of being counted as an error: "nothing to extract" is a valid scan result.

Smarter Agent Mode assistant

The Agent Mode assistant now knows Claira's full feature set: ask "what are Insights?" or "what is a bulk scan?" and it explains the feature in plain language first, then offers "Do you want to learn more?" instead of jumping straight into setting up a scan. See Agent Mode.
The assistant can now search the help site for you: accept its learn-more offer (or simply ask for documentation) and it searches the support articles and replies with the most relevant links. Help-article matching is much more reliable than before, including accent-insensitive French search.
Ambiguous terms now default to their Claira meaning: "scans", "fields", "tokens", "insights" and similar words are read as Claira features unless your question is clearly unrelated to the product.
The "What can Claira do" overview (the chat's starter answer and the article) now also covers Insights and Field Migration.

Plan changes now happen only in Settings

The Allow auto upgrade case setting has been removed, along with the one-click Upgrade Now buttons in out-of-capacity popups. When a scan or chat request is blocked, the popup now shows what is blocking it, including the exact tokens needed vs. available, and its Open plan manager button takes you to the Settings plan manager, where all plan changes are made manually. Running past your balance is handled by overage billing (Allow overages), so nothing changes your plan or your bill without an explicit change in the plan manager. The Agent Mode View plans / Upgrade chat button is now View plans and opens the same plan manager. See Plans and Billing: Overage.

Rate responses and Smart Revise

Completed AI responses now carry small feedback buttons at their bottom-left corner: thumbs up and thumbs down in the Prompt Lab, the Multi-Code single-document view, and Agent chat, plus a new Smart Revise wand on document scans. See Rate responses and Smart Revise.
Smart Revise turns a comment ("the summary should list key dates as bullets") into ready-to-apply prompt edits. On Multi-Code runs it works out which field's prompt your comment concerns (or revises the shared general instruction) and one click applies the revision to the right editor. It's free and never re-runs anything automatically.
Hovering any feedback icon explains what it does and notes that the response and document data are shared with the Claira team, who review feedback weekly to improve Claira.

"Use Multi-Code" splits prompts properly, and fails safely

Use Multi-Code in Prompt Lab now writes real instructions into each field prompt instead of bare labels. When a prompt ends with an output-format block (RELEVANCE: [YESREL/NOREL], JUSTIFICATION: [1-2 sentences]), Claira reads those labels as the names of the outputs and moves the matching definitions and criteria out of the body of your prompt into each field's own prompt, so every field can decide its answer on its own, instead of everything being left in the shared instruction. See Objective coding: Multi-Code mode.
Each converted row is now named after its output, a destination field you had already mapped is kept when the name still matches, and an unmapped row is pre-filled when a case field has exactly that name, so you no longer have to re-pick every destination after a conversion.
If a converted field prompt still looks too thin to decide an answer, Claira names those fields in a banner on the Multi-Code card and asks you to review them before scanning, instead of reporting the conversion as a clean success.
If the conversion cannot be completed (Claira is momentarily busy, or the prompt genuinely does not split into separate fields), nothing is changed: your prompt stays exactly as you wrote it, the message tells you which of the two happened, and you get a Try again button. Previously a failed conversion still switched you into Multi-Code, where you would find an empty configuration and have to go back for your prompt.

Create in Discover always available

The Create in Discover tab no longer greys out with "This Discover portal doesn't support creating fields from Claira yet", a pre-check that was wrongly blocking field creation on many cases where it works fine. Everyone now gets the create form (and the agent can always propose fields); on the rare portal that genuinely can't create fields, the attempt fails safely with that same clear message and nothing is created. Permission problems also report themselves more precisely: if your Discover role can't create fields, you get the "Your Discover role may not allow creating fields" message instead of a generic error. See Creating fields: Troubleshooting.

Field Migration: value matching, memo splitting, pinned fields, and date fixes

Migrating into a pick list no longer requires exact matches: a new matching table lists every distinct source value in your selection and lets you pair each one with a destination choice, or skip it. Exact name matches are paired automatically. See Matching values to pick-list choices.
A new Split each value on a delimiter option for memo sources turns a value like smith@x.com; jones@y.com into separate destination values using any separator you type. The destination must accept multiple values: the option explains itself right there when the chosen destination can't. See Field Migration: Options.
The fields Claira recognizes as your case's Date and Title fields now appear pinned in a Recommended group at the top of both field pickers. See Field Migration: Choose your fields.
Fixed date-to-date migrations that could wrongly skip documents whose dates Nuix Discover returns with a time component, and fixed undoing a test on a date field. Migration results are also stricter about what counts as migrated: when Discover reports back that a write did not take effect, those documents are now marked failed instead of counted as migrated.

Insights: sturdier citations, partially-completed scans as sources, and visible references

Citation checking no longer depends on the AI writing its references in one exact format: bracketed document IDs are now recognized as citations whenever every ID in the brackets is one of the Insight's source documents, in the viewer, in the PDF export, and in the stored Insight itself. Ordinary bracketed text like "[sic]" is never mistaken for a citation, and an ID that is not among the source documents is still flagged in red. Existing Insights benefit too, without regenerating. See Insights.
A bulk scan that finished with some failed documents can now be used as an Insight source. Previously a single failed document out of thousands silently removed the whole scan from the source picker. Only documents with a completed result feed the Insight, and when you pick such a task, a note under the picker shows exactly how many will be included (for example, "113 of 115").
Every Insight now shows its reference (IN + number, e.g. IN012) on its row in Past insights and in the viewer's header, so two insights with similar titles are always tellable apart, and typing a reference in the search box finds it. It is the same reference the Log Field stamps and the PDF header prints.

Release 053 -- July 18, 2026

Claira now creates Discover fields for you end to end, background scans hand off cleanly with a live colour-coded progress bar, bulk-scan billing is corrected in every mode, and the French interface gets a full accent polish.

Auto scan mode fixes

Fixed the bulk scan confirmation dialog showing "Extracted text" as the scan input mode when Scan as was set to Auto. It now shows "Auto (best type per document)". The scan itself always ran in the selected mode; only the dialog label was wrong.
Fixed the Auto scan mode choice not being remembered across browsers and devices: switching back to Auto after picking another mode is now saved with your case settings like the other scan modes.

Generating Search Term Families workflow article

Added a new workflow article, Generating Search Term Families: a paste-in instruction block that teaches Claira Nuix Discover (dtSearch) search syntax so it outputs term queries ready to import as search term families, with step-by-step import instructions and example output.

Create Discover fields from inside Claira

Claira can now create working fields directly in Nuix Discover, with no more looking up composite IDs by hand. The field dialog gained a Create in Discover tab: pick a name and a type (Memo, Text, Number, Date, or Yes/No) and Claira creates the field and connects it automatically. The previous paste-a-composite-ID flow lives on as the Connect existing tab.
Created fields are connected and selectable immediately. Discover creates new fields without group access, so right after creation Claira shows copy-ready instructions for your case administrator: have read and write granted in Discover before scan results can be written to the field. If a scan later can't write to it, the missing grant is the reason.
Claira's Agent can now suggest and create fields in chat: it proposes a name and type on a confirmation card (you can edit both before creating), reminds you of the admin access step right on the card, and can plug the new field straight into a waiting bulk scan. Because Discover fields can't be deleted or renamed, nothing is ever created without your explicit click on the card.

Clearer messaging when starting a background bulk scan

Starting a background bulk scan no longer shows the (always-empty) live results panel. Instead, Claira shows an Assigning your Task ID… notice asking you to keep Claira open for the few moments of hand-off, then a green Task #NNN is running in the background confirmation the instant it's safe to close your browser, with the task number, so you can find the run in Ongoing background tasks later.

Bulk-scan billing and reliability fixes

Fixed regular (single-field) bulk scans being billed at the Multi-Code rate: a text scan charged 2 tokens per document instead of 1 (image 6 instead of 5, and so on). Regular bulk scans are billed at the base per-document rate again; Multi-Code keeps its flat base-plus-one pricing. The same over-count also made the pre-scan eligibility check double the estimated cost, so it could block scans your balance actually covered.
Fixed live bulk scans sometimes stopping early with "The scan stopped before finishing…" on selections containing many documents with no readable text (or text over the size limit). A skipped document could silently halt one of the scan's parallel workers; with enough skipped documents the whole run stalled. Skipped documents are now counted and the scan carries on to the end of the selection.
When a bulk scan is blocked because the case doesn't have enough tokens (and overages are off), the message now shows the exact numbers (how many tokens the scan needs and how many the case has available) instead of a bare "insufficient balance" notice, and points to the Allow overages switch under Settings → Manage plan.

Background progress bar: counts in the bar, a legend, and live skipped documents

The background-task progress bar now shows each phase's document count in white inside its colored segment (with a tooltip when a segment is too narrow), and a color legend at the top of the Ongoing background tasks list explains every color in pipeline order.
Documents skipped during a background run (no readable text, or text over the size limit) now show live in their own pink segment while the run is still going, instead of sitting in the amber "extracting" color until the end.

Scheduled downgrades are now visible and cancellable

Scheduling a plan downgrade now shows in Settings → Manage plan: the target plan's card reads Downgrade scheduled on the effective date, your current plan's card notes the upcoming change, and a Cancel downgrade button undoes it at any time before it takes effect. You keep your current plan's full capacity until the effective date. This also works for credit-card subscriptions, whose scheduled downgrades were previously invisible.

Video markers on help articles

Help articles that include a walkthrough video now show a small video icon next to their link (in the navigation sidebar, in links from other articles, and on the Workflows index), so you can tell at a glance which articles have a video before you open them.

Author and Recipients templates for objective coding

Added two new prompt templates under Objective coding: Extract Author and Extract Recipients. They name the individual and/or organisation a document is from and to: the primary sender or creator (email sender, invoicing company, letter writer) and the direct recipients (never Cc or Bcc), in a consistent format suited to affidavit-style coding.
Renamed the email-address templates so their purpose is clear at a glance: Email To, Email From, Email Cc, and Email Bcc (formerly "Extract To/From/Cc/Bcc Addresses"). The prompts themselves are unchanged.
Retired the Extract People template: Extract Author does the same job across more document types. Links to the old help article now land on Extract Author.

Release Videos section

Added a new Release Videos section to the help desk: short video tours of what's new in each major Claira release, added every few weeks alongside the release. The first entry is the July 2026 Update, which walks through Field Migration, shared Agent Mode chats, background Multi-Code scans, flat Multi-Code pricing, and higher bulk-scan limits.

Summarize Document template: one prompt for every scan mode

The Summarize Document template now inserts the same prompt whether you scan documents as Text, Image, Audio, or Video. It previously swapped in separate wording for image/raw-file scans; a single, mode-neutral prompt now covers every case, so you get consistent results regardless of how a document is scanned.

French template details now display in full

Fixed the French prompt template picker showing blank Step-by-step, Recommended customizations, Troubleshooting, and Related sections when you expanded a template. All four now show their full content, matching the English picker.

Multi-Code auto-mode billing fix

Fixed bulk scans with Multi-Code enabled and Scan as set to Auto not charging the flat 1-token Multi-Code surcharge: the end-of-task refund returned the surcharge for every successful document, so these runs were billed at the plain scan rate instead of the documented Multi-Code rate (scan rate + 1 per document). Successful documents in these runs are now billed exactly as the pricing table states; failed, skipped, and unprocessed documents are still fully refunded.

Smoother completion toasts and truthful token numbers

The bulk-scan completion toast now dismisses itself after about 15 seconds and appears below the header bar, so it no longer covers the Menu button or the top-right controls of the task detail window while you review results.
Task details for Auto scans now show the tokens actually charged after end-of-task settlement instead of the worst-case hold (a 4-document Multi-Code Auto run that settled to 24 tokens used to display 84).
Live scans and background scans now count documents the same way: a document whose text can't be fetched at all is recorded as Skipped in both run modes, instead of "Failed" in live runs only.
Bulk tasks that never started (for example when the browser was closed during the hand-off) are now cleaned up automatically and their held tokens refunded, and they no longer block disconnecting a field they referenced.

French interface accents restored

Restored missing accents across the French interface: the plan manager and upgrade/blocked-scan dialogs (« Gérer le forfait », « Mise à niveau requise »…), the template guide section headings (« Personnalisations recommandées », « Dépannage »…), the « Jetons utilisés » task statistic, and a number of other strings.

Release 052 -- July 9, 2026

The new Field Migration tool copies and moves field values, Agent Mode chats become a shared team workspace, Multi-Code scans run in the background at a flat +1-token price, and per-plan bulk-scan caps are gone.

Highlights

Choose your scan speed, mid-run. Live bulk scans now show a Scan speed slider in the progress area, with five positions from Slowest to Fastest. Slower positions all but eliminate resource-availability errors; faster positions finish sooner but are more likely to run into them during busy hours. You can move the slider at any time while the scan runs: changes take effect immediately, and the estimated duration updates to match. See Bulk Scan: Adjusting the scan speed.
A calmer default. Every live bulk scan now starts at the Balanced speed, which runs a little below Claira's maximum, the sweet spot that keeps resource-availability errors rare. When your run looks clean and you want results sooner, push the slider toward Fastest to reclaim full speed. The guide How to speed up your bulk tasks covers when to do so.
Replenishment popup centered at every window size. The animated popup that appears when your token reservoir refills while you were away (including the "Token reservoir fully recharged" celebration) now always appears centered at the bottom of the Claira panel. Previously, on smaller windows the full-reservoir version appeared shifted to the right, sometimes partly out of view, and skipped its entry animation; the falling-token droplet and splash now line up with the bucket as well.
Multi-Code bulk scans can now run as background processing tasks. On cases with Background Processing set up, the Run as background processing task checkbox now appears for Multi-Code text scans too: Claira codes all of your configured fields for every document server-side, emails you when the run finishes, and you can close Nuix Discover while it works. See Bulk Scan for details.
Agent chats are now shared with your case team by default. Teammates can open and read your shared chats (never edit them), each chat shows its owner in the sidebar, and a toggle in the chat's top bar switches any chat between Shared with case team and Private to me. Claira remembers your last choice for new chats, and all existing chats became shared. See Sharing, pins, and branching.
Branch a shared chat: continue a teammate's conversation in your own copy. The original stays untouched and your copy keeps a reference to where it came from. You can branch your own chats too.
Pin agent chats for yourself or for the whole team (up to 5 team pins per case), exactly like pinned prompts: pinned chats float to a Pinned section at the top of the sidebar.
Filter the chat list by user (Everyone / Me / specific teammates), matching the bulk scan history filter, and search now also matches owner names.
The chat sidebar got a cleaner, more compact design: slimmer rows with the last-activity date, and full dark-mode support. Details in the updated Agent Mode article.
The Bulk Scan Task ID field is now the Log Field. The same connected field now records a reference for every Claira operation that touches a document, not just bulk scans: bulk scan tasks write SC + the task number (e.g. SC067), and Insights write IN + the insight number (e.g. IN012). New operation types can be added in the future without connecting any new fields. See Fields.
Insights now tag their source documents. When you create an Insight, every document whose results feed into it receives the Insight's reference (e.g. IN012) in the Log Field, so you can pull the exact source set up in Nuix Discover with a single search. The Show sources panel and the Insight PDF show the reference to search for. See Insights.
Task details always show the exact stamped value. Tasks that ran before this release were stamped with the bare task number (e.g. 067); their details window and search instructions keep showing that historical value, while new tasks show the prefixed form.
New: Field Migration. Copy or move values from one Nuix Discover field to another across your current document selection, with copy/move and skip/overwrite options, a safe one-document test with undo, live progress, and a browsable migration history with per-document results. Each run tags migrated documents with its MG### reference in your case's Log Field. Free, with no tokens used. See Field Migration.
The mode menu now has a More section at the bottom (hover to open): Field Migration, Multi-Code, Case Context, and Settings live there, keeping the main menu focused on everyday modes.
Multi-Code bulk scans now cost a flat +1 token per document, whatever the number of fields. A bulk Multi-Code scan now adds a single token per document on top of the base scan rate (2 tokens for text, 6 for image, 11 for audio, and 21 for video per document), whether you configure a single coding field or all eight. Previously the cost grew with each field you added. Running Multi-Code on a single document in review stays free. See Bulk Scan.
Every plan can now bulk-scan up to 999,999 documents at a time. The per-plan cap on how many documents a single bulk scan could include has been removed: Starter, Pro, Business, and Enterprise now all allow selections of up to 999,999 documents in one task, for every bulk scan type (including Multi-Code). Your token balance still applies, so large runs need enough tokens (or overage enabled) to complete. See Bulk Scan and Plans and billing.
If a selection ever exceeds the 999,999-document maximum, Claira now simply asks you to reduce it, instead of prompting an upgrade. The maximum is the same on every plan, so upgrading does not raise it.

Release 051 -- July 7, 2026

Agent Mode threads name themselves, the bulk scan confirmation leads with cost, and Multi-Code live results are now readable field-by-field.

Highlights

AI-generated thread titles. When you send the first message in an Agent Mode thread, Claira generates a concise title (about 5-7 words) summarizing your question and shows it in the thread sidebar, the same naming experience as bulk scan tasks and Insights. Titles are written in the language of your question (English or French). Until the title is ready, the sidebar briefly shows the beginning of your question. See Agent Mode.
Cost decision first. The bulk scan confirmation dialog now opens with a Usage and cost section showing the estimated token usage and your available token balance side by side, directly above a confirmation that the task fits within your balance, or a notice with the potentially billable token count when it doesn't. See Bulk Scan: How to run a bulk scan.
A tidier summary. The estimated duration now appears as a row in the task summary next to the document count and scan input mode, instead of in its own section at the bottom of the dialog, and the remaining sections share one consistent layout.
No contradictory warning on background runs. The Keep this case open warning now appears only for foreground runs. When Run as background processing task is checked, the warning disappears: background runs keep working after Nuix Discover closes.
Multi-Code entries formatted like a single review. During a Multi-Code bulk scan, each entry in the live results panel now shows the extracted values as Field: value lines, one per destination field, matching how a single-document Multi-Code response is presented. Previously the feed showed the raw JSON payload returned by the model. See Bulk Scan: Monitoring progress.

Release 050 -- July 6, 2026

Background scans now show a color-coded progress bar, write results back as they go, and ask what to do with finished work when you cancel.

Highlights

A stacked progress bar for background scans. The ongoing-task card now shows a color-coded bar that fills from left to right as the run advances: green for results written back to Nuix Discover, teal for AI processing complete and waiting to be written back, blue for submitted for AI processing, and yellow for text extraction, replacing the previous "Step N of 3" line. Beneath the bar, per-phase count lines appear only while they're informative. See Bulk Scan: Monitoring background tasks.
Review can start before the task finishes. Results are now written back to Nuix Discover progressively while a background task is still running (including very large scans), so you can begin reviewing completed documents right away instead of waiting for the whole run. See Bulk Scan: Where results go.
Cancelling a background task now asks about finished work. Clicking Cancel task opens a dialog that tells you roughly how many documents have finished processing and offers three choices: Write back results & cancel (completed results are saved to the destination field, and you are charged only for the documents actually written), Discard & cancel (permanent; the tokens reserved for unwritten documents return to your balance), or Keep running. No email is sent for a cancelled task. See Bulk Scan: If a scan stops early.
Completed work survives an unattended failure. If a background scan hits a failure it cannot recover from while nobody is watching, Claira automatically writes back everything that completed first, the notification reports honest counts of what was written and what was not, and documents that could not be completed get an explanatory ERROR_LOG note in the destination field.
Very long runs keep themselves going. Long-running background scans are now continued automatically behind the scenes: the task keeps going and the bar keeps advancing, with nothing for you to restart. See Setting Up Background Processing: Running a background scan.

Release 049 -- July 6, 2026

Every plan now includes every Claira feature: plans differ only in token capacity.

Highlights

Multi-Code on every plan. Multi-Code (multi-field extraction) no longer requires Pro or higher: Starter cases can run single-document and bulk Multi-Code, paying the same per-field token costs as every other plan. See Plans and Billing and Single review.
Case Context on every plan. Building and using Case Context is now available on Starter as well.
Unlimited saved Agent Mode threads. The Starter limit of 10 saved threads per user per case is gone. Every plan keeps as many threads as you need. See Agent Mode.
One simple rule. Media scans (image, audio, video), Auto mode, background runs, Insights, and Agent Mode were already open to all plans; with this release nothing in Claira is gated by plan anymore. Plans differ only in price and token capacity: reservoir size, daily refill, and bulk task limit. See Plans and Billing.

Release 048 -- July 6, 2026

Deactivating a case and deleting its data are now one connected flow: deactivate at period end and keep the data, or deactivate now and delete everything.

Highlights

One Deactivate Case action, two clear choices. The Deactivate button now lives in the Settings Danger Zone and opens a confirmation dialog with two options: Deactivate only keeps Claira available until the end of the current billing period (with the date shown and a cancel button while it's pending), or Deactivate now and delete all case data, which takes effect immediately and schedules permanent deletion after the standard 7-day grace period. See Deleting a case's data: Deactivating a case.
Deleting data no longer requires a support request. Previously, a deactivated case could never reach the delete-data action on its own. Now deletion unlocks as soon as the case is deactivated, including from the case's activation screen, which shows a Previously active badge (hover for the deactivation date) and a Delete all case data permanently option for inactive cases. See Deleting a case's data: Deleting data from a deactivated case.
Credit-card subscriptions end with the case. Deactivating now also ends the case's credit-card subscription, at the period boundary for deactivate-only, or immediately (without a refund) when deleting, so a deactivated case can no longer keep billing.
Reactivating keeps your data safe. Choosing a plan again on a previously-active case automatically cancels any pending data deletion, and the activation screen shows the deletion countdown with its own cancel button during the grace period. See Deleting a case's data: Reactivating a case.
Everyone stays informed. The case's email recipients receive a confirmation when a deactivation is scheduled (with the effective date) and a single combined email when a case is deactivated with deletion (with the deletion deadline).

Release 047 -- July 6, 2026

Disconnecting a field is now blocked only by an active bulk scan: teammates' selections no longer stand in the way.

Highlights

Disconnect a field at any time. Previously a field could not be disconnected while anyone on the case had it selected in their scan settings. Now the only thing that blocks a disconnect is a bulk scan that is running or queued and writes to that field. Finish or cancel the scan, then disconnect. See Fields: Disconnecting fields.
Teammates' selections update themselves. When you disconnect a field that others had selected, their selection is cleared automatically: they simply pick a different field the next time they run a scan.
Scan history stays intact. Disconnecting removes only the connection between Claira and the field. The Nuix Discover field and every AI response already written to it are untouched, and past scan results and history in Claira are unaffected. Reconnect the same field later and it picks up right where it left off.

Release 046 -- July 6, 2026

Insight and bulk-task titles are now click-to-edit, with a one-click AI regenerate icon next to every title.

Highlights

Rename an Insight any time. Click the title at the top of the Insight viewer, type the new name, and press Enter. Until now Insight titles could not be changed at all: if automatic naming failed, the placeholder stayed forever. See Insights: Renaming an Insight and regenerating its title.
Regenerate any title with AI. A small refresh icon now sits next to the title in both the Insight viewer and the bulk task details. Hover it and the tooltip reads Regenerate title with AI. One click asks Claira to propose a fresh title from your original prompt. On bulk tasks this replaces the old Regenerate Name button that only appeared when a name was missing or cut off. See Bulk Scan: Task names.
Editing a name now takes a single click. Task names in the Bulk Tasks list and task details (and now Insight titles) open for editing on a single click instead of a double-click.

Release 045 -- July 6, 2026

A refresh icon on the activation screen re-checks a case's activation status on demand.

Highlights

Re-check activation without reloading. If the activation screen appears for a case you know is already activated (for example after a brief connection hiccup while Claira starts up), click the new refresh icon next to the activation title to re-check. If the case is active, the screen closes by itself and Claira loads normally. See Case Activation: If this screen appears on an already-activated case.
Background-scan result emails now identify the run at a glance. The subject line leads with [Claira], the task name and number, and the case name, and the email body adds the final status, how long the scan took, and the field the responses were written back to, so if you're juggling several scans across cases, you can tell which run finished and how it went without opening the system. The document counts (successful, skipped, errored) are unchanged, and the failure email now carries the same task and case identification. See Setting Up Background Processing: Running a background scan.

Release 044 -- June 12, 2026

A bulk scan that stops early is now marked Incomplete, with one-click Resume, instead of showing as Completed.

Highlights

Honest task status for interrupted scans. If a bulk scan stops before every document has an outcome (closed tab, dropped connection, a burst of AI-service errors), the task now shows an amber Incomplete badge instead of "Completed". The task details tell you exactly how many documents still need scanning.
Resume picks up where the scan left off. Click Resume Task on an Incomplete task to scan only the remaining documents: documents that already have results are not re-scanned and are not charged again. Prefer not to finish? Cancel task releases the reserved tokens back to your balance.
Counts you can trust. When a task finishes, the Completed / Failed / Skipped counts are now reconciled against the per-document records, so the numbers in the task details and the Bulk Task History Report always add up to the documents you selected.

See Bulk Scan: If a scan stops early and the updated Troubleshooting: Bulk scan stopped or failed for details.

Release 043 -- June 12, 2026

Background scans now show step-by-step progress and finish on their own, at any size.

Highlights

Step-by-step progress for background scans. While a background bulk scan runs, the Bulk Tasks list now shows exactly where it is: Step 1 of 3 · Extracting text · 400 of 5,000, then processing with AI, then writing results to Discover, with live document counts that update every time you refresh the view. See Bulk scan: Monitoring background tasks.
Background scans of any size finish on their own. The background engine now works in self-contained shifts that pick up exactly where the last one ended, so very large tasks (up to the plan maximum) no longer need babysitting: interruptions, restarts, and long waits are absorbed silently and the run simply continues. A failure email now means the run genuinely stopped making progress and needs your attention; everything else heals itself. Previously, a long run could be reported as failed while its work actually finished fine, asking you to resume a task that didn't need it.

Release 042 -- June 11, 2026

Choose who receives Claira's automatic emails, and monthly invoicing now covers your whole organization.

Highlights

Email preferences in Settings. A new Email preferences section in Settings lets your team decide, per email type, whether Claira's automatic emails go to everyone in the case or only the person who activated Claira. It covers activation, plan-change, token-reservoir, and background-scan emails, and the choice is shared across the case: anyone can change it, and it saves instantly. Defaults match the previous behavior, so nothing changes until you adjust them. See Email Preferences.
The "Email only me" activation choice is now remembered. Checking Email only me when activating a case sets every email type to only the activator; leaving it unchecked keeps the standard defaults. Either way, the choices can be changed later from Settings. See Case Activation Broadcast.
No more per-organization approval for monthly invoicing. Once monthly invoicing is enabled for your organization's billing relationship, every Nuix Discover organization working under your Claira configuration can activate cases with it, including organizations created after your initial setup. Previously, each Nuix organization had to be approved individually, so teams in a new sub-organization had to contact support before they could choose monthly invoicing. Credit-card billing is unaffected and remains available to everyone. See Plans and Billing and How monthly invoice billing works.

Release 040 -- June 10, 2026

Background scan results now display their formatting.

Highlights

Formatted results in background runs. When a bulk scan runs as a background processing task and writes to a long-text response field, the AI's formatting (bold and italic emphasis, bullet and numbered lists, headings, and line breaks) now displays properly in Nuix Discover, exactly as it does for foreground scans. Previously, background results could show raw symbols like **bold** or - item, and multi-line answers could run together on a single line. Nothing changes about how you launch a scan, and ERROR_LOG entries keep their plain-text format. See Bulk scan for how bulk scans write results.

Release 039 -- June 9, 2026

Background Processing is now available to everyone.

Highlights

Run bulk scans in the background. The Run as background processing task option now appears in the bulk-scan confirmation popup in every case that has a background service account configured, with no need to keep Nuix Discover open while a long scan runs; Claira finishes the work server-side and emails you when it's done. The engine behind it was rebuilt for reliability: documents are fetched and written back in small, individually safeguarded groups, so one problem document or a brief service interruption no longer affects the rest of the run, and an interrupted task resumes exactly where it left off. Background runs are deliberately steadier, and slower, than foreground runs. Available for single-code text scans on Gemini models.
Every document gets a result, including the ones that couldn't be scanned. A background run now writes an explanation to the response field the moment a document is ruled out, using the same ERROR_LOG: [E-…] markers as foreground scans: documents with no extracted text or whose text could not be retrieved are marked [E-FETCH], oversized documents [E-TOOLARGE], AI errors [E-AI] or [E-SAFETY], and write failures [E-WRITE]. No document finishes a background run with an empty response field, and every marker is explained in Error codes.

See Background Processing setup to connect the service account that enables the option.

Release 038 -- June 9, 2026

Choose who gets the activation email: everyone in the case, or just you.

Highlights

"Email only me" on case activation. When you activate Claira in a case, every case user normally receives a short email letting them know Claira is now available there, a simple way for the whole team to discover the new capability. A subtle new Email only me checkbox just above the Activate Claira button lets you keep that announcement to yourself instead: check it and the activation email goes only to you. It is unchecked by default, and an info icon next to it explains the choice.

See Case Activation and Case Activation Broadcast for details.

Release 037 -- June 9, 2026

The Starter plan now includes a daily token refill of 20 tokens.

Highlights

Daily refill on Starter. Starter cases now receive 20 tokens every day, automatically topping your balance up toward the 1,000-token reservoir cap, the same daily-refill mechanics Pro and Business plans already use. Previously, Starter's 1,000 tokens were granted once at subscription with no refill. Nothing changes about the price or the reservoir cap, and the refill applies to existing Starter cases as well as new ones. See Plans & Billing for the updated plan details and Usage analytics for how refills appear in your token ledger.

Release 036 -- June 9, 2026

Every help page now ends with a quick way to reach support.

Highlights

Support contact on every page. Every article on this help site now ends with a "Need more help?" card showing our support address, support@claira.to, in both English and French. If a page doesn't answer your question, the way to reach us is right at the bottom, and the email opens with the article name prefilled in the subject line, so we know which page you were reading.
The language toggle now always lands on the matching article. On a handful of help pages, switching languages with the "Voir en français" / "See in English" button could show the English article at the French address. The toggle now takes you to the proper translated page on every article.

Release 035 -- June 5, 2026

Bulk scan progress now counts skipped documents separately, so "processed" matches the results you see.

Highlights

A live "skipped" count during bulk scans. While a scan runs, the progress counters now show processed, skipped, and failed as three separate totals. "Processed" now reflects only the documents that produced a result (the same documents listed in the live results panel), while documents with no readable text or media (or whose text is over the size limit) are tallied under "skipped" instead of being counted as processed. Failed documents are unchanged.

See Bulk Scan: Monitoring progress for details.

Release 034 -- June 4, 2026

Hover the live results during a bulk scan to pause the feed and read an entry without it scrolling away.

Highlights

Pause-on-hover for live results. While a bulk scan is running, moving your cursor onto the live results list now pauses the stream, so an entry you've expanded to read stays in place instead of being pushed down as new results arrive. Results keep coming in while you read and appear the moment you move your cursor away, so nothing is missed.

See Bulk Scan: Monitoring progress for details.

Release 033 -- June 4, 2026

The daily token-reservoir emails are friendlier and now name your case.

Highlights

A full reservoir reads as good news. When a case's token reservoir reaches its full capacity, the email now leads with the good news, names the case in the subject and body, and shows how many tokens are available, using the same styling as Claira's other billing emails. It replaces the old terse wording that looked like an error.
The daily top-up email matches. The daily "tokens added" summary uses the same clean styling, names the case, and shows both the tokens added and your current balance.

See Email Notifications Overview for details.

Release 032 -- June 3, 2026

Very large documents are skipped cleanly during a scan instead of interrupting it.

Highlights

A single oversized document can't derail a scan. When a document's extracted text is too large to scan (about 3 MB or more; typically a huge spreadsheet export, a long transcript, or a heavily OCR'd file), Claira now skips it with the new E-TOOLARGE result and keeps going. The skipped document is marked skipped, not failed, and isn't sent to the AI model. Split it with Unitize in Nuix Discover to scan it in smaller parts.
Steadier large bulk scans. Foreground bulk scans handle cases full of large technical documents more reliably, with lower in-browser memory use.

See Error codes: E-TOOLARGE for details.

Release 031 -- June 2, 2026

A new billing ledger at the bottom of Usage mode shows your case's billing history at a glance.

Highlights

One newest-first ledger of billing activity. Usage mode now ends with a colour-coded billing ledger: plan upgrades and downgrades, monthly subscription payments, bulk scans, token purchases, overages, refunds, and Insight revisions, each with its date and amount. Token movements and dollar amounts have their own columns, and a token purchase shows both.
Filter and focus. Narrow the ledger to a single event type, and reveal the daily token-replenishment rows with a checkbox (hidden by default so the list stays readable). The ledger pages 20 events at a time, newest first.
Monthly payments on their billing date. Each recurring subscription charge now appears in the ledger on the date it was paid, linking to its invoice where available.

See Usage Analytics: Billing ledger for details.

Release 029 -- May 28, 2026

GPT-OSS 120B is available again for Canada tenants.

Highlights

New model option in the AI Settings picker for Canada tenants. GPT-OSS 120B is back as a third option alongside Gemini Flash and Gemini Pro 2.5. The model is hosted inside the Canadian data residency boundary and only appears in the picker for tenants whose deployment is in Canada. Australia tenants continue to see Gemini Flash, Gemini Pro 2.5, and Claude Sonnet 4.6. See Model selection: Available models for when to pick it.

Release 027 -- May 26, 2026

Monthly invoice billing doc now covers overages and uses plan names instead of dollar amounts.

Highlights

New Overages section in the monthly invoice article. How monthly invoice billing works: Overages explains how overage accrues on invoice accounts when a period runs through its allotment, why it is collected on your invoice date rather than your plan date, and how that timing interacts with mid-cycle plan changes and cancellation. The section links back to Plans and Billing: Overage for the per-token rate and case-settings detail.
Worked example uses plan names instead of specific dollar amounts. The example in How monthly invoice billing works: A worked example now refers to the Business and Pro plans by name, so the timing logic stays correct as plan prices evolve.

Release 026 -- May 24, 2026

Auto mode now always sends the source file when one exists, including every PDF.

Highlights

Auto picks the source file whenever the model can read it directly. When a document has a PDF, image, audio, or video file attached in Nuix Discover, Auto now sends that file instead of the extracted text. The previous distinction between "scanned PDF" (sent as image) and "digital-native PDF" (sent as text) has been removed: PDFs are always sent as the PDF itself, so the model sees layout, signatures, redactions, and handwriting that an OCR layer may have missed. See Media scans → How Auto picks per document.
Text-mode fallback is now reserved for documents without a native file. Word documents, Excel sheets, emails, HTML, plain .txt, and other formats the model can't ingest directly continue to be reviewed against the extracted text. Oversize media files still fall back to text when extracted text is available.
Pre-scan indicator on single scans. When Auto is selected for a single-document scan in Single Review or Multi-Code, the active document now shows "Will scan as: Text / Image / Audio / Video" next to the Scan button before you click, so you know which mode the run will use without guessing.
Bug fix: Auto-mode single scans were silently running as text. A GraphQL query in the single-scan resolver requested an extracted-text field with the wrong shape, causing every Auto scan to fall back to text regardless of the actual file type. Auto now correctly picks the native media mode for video, audio, image, and PDF documents.

Release 025 -- May 24, 2026

New AI review article: how to speed up bulk tasks.

Highlights

New documentation page. How to speed up your bulk tasks gives reviewers four practical levers for shorter bulk runs: keep prompts short, match the model to the task (Fast vs Smart), trim the Nuix Discover selection before launching, and run very large jobs at off-peak times. The article cross-links to Model selection, Single review, and Prompting.

Release 024 -- May 24, 2026

Self-serve scheduled downgrades on monthly invoice plans.

Highlights

Downgrades on invoice accounts are now self-serve and scheduled. Invoice accounts can downgrade their plan tier in the plan manager without contacting support. Claira confirms the request immediately and schedules the change for your next plan date, so you keep the larger allotment you paid for through the end of the current period; the lower plan begins automatically when the period ends. See How monthly invoice billing works: Mid-cycle plan changes.
Cancel a scheduled downgrade implicitly. Requesting your current tier in the plan manager clears any pending scheduled downgrade. Requesting an upgrade also overrides a pending downgrade and takes effect immediately.
Two transactional emails for scheduled downgrades. When you schedule a downgrade you get a confirmation email naming the date the change takes effect; when the deferred change applies, you (and other case admins) get a follow-up email confirming the new plan is now active.

Release 023 -- May 24, 2026

New admin article: how monthly invoice billing works.

Highlights

New documentation page. How monthly invoice billing works walks accounts on the monthly invoice option through the two dates that govern their billing (plan date vs invoice date), why each period carries one full plan charge with no proration, how mid-cycle upgrades and downgrades are handled, and what happens when a plan is cancelled. The article includes a worked example and a short comparison with credit-card billing. Plans and Billing now links to it from the contractual-invoicing section.

Release 022 -- May 24, 2026

Small improvements to prompt history, bulk scans, and Multi-Code.

Highlights

Pinned prompts in prompt history. You can now pin a prompt from your prompt history so it stays at the top of the list. Pins have two scopes: Just me (only you see them) and Everyone on this case (visible to every member of the case, capped at 5 per case). Case pins appear above personal pins, each ordered by most-recently-pinned.
Live results feed during bulk scans. While a bulk scan is running, a new panel inside the active-scan card streams completed documents as they finish, so results are visible as they land instead of only when the full run completes. Failed and skipped documents continue to roll up into the counter cards above.
Multi-Code: partial-success writes. If one or more fields fail on a document during a Multi-Code scan, Claira now writes the fields that succeeded and only marks the document as failed when every field fails. Previously a single-field failure would void the document's entire output.
Prompt log on each document names the destination field. When Claira writes the prompt log to a document during a scan, the memo now starts with a bold line naming the response field the prompt was written to, followed by the prompt text. Reviewers reading a document's prompt log can see which field a given prompt produced without opening the bulk task or prompt history.

Release 021 -- May 24, 2026

Auto media-type mode arrives, and every scan mode is now available on every plan.

Highlights

New "Auto" option in the Scan as dropdown, and it's the new default. Claira inspects each document at scan time and picks the richest mode that fits: Video on a video file, Audio on an audio file, Image on a scanned PDF, and Text on everything else (Word, Excel, email, native PDFs, etc.). Mixed productions are no longer a problem: the same Auto run produces the right treatment per document, with no pre-sorting on your end. See Media scans for the full hierarchy.
Oversize fallback so scans don't error out. If the chosen media file exceeds the 50 MB per-document size cap and the document also has extracted text available, Auto falls back to Text rather than skipping. If a document has neither readable media nor text, Auto marks it as skipped and continues with the rest of the run; the reason is recorded on the document in the task history.
All media scans (Image, Audio, Video) are now available on every plan, single and bulk. The Pro-tier gate on media scans has been removed: Starter customers can run any mode without an upgrade prompt. Token costs (5 / 10 / 20 per doc) still apply and the reservoir cap on your plan is what limits volume.
Per-document billing for Auto bulk runs. A bulk Auto run reserves the worst-case rate (20 tokens / doc) up front and refunds the difference as each document resolves to a cheaper mode. The history page records the actual mode Auto picked per document, so you can audit the mix and the resulting cost after a run completes.
Existing dropdown preferences are kept. If you've already picked Text, Image, Audio, or Video at some point, your selection is preserved: Auto only becomes the default when you've never picked. You can switch to or away from Auto at any time.

Release 020 -- May 24, 2026

Self-serve case data deletion after plan cancellation.

Highlights

New "Danger zone" in case Settings. Once a case's plan is cancelled, a new section at the bottom of the case Settings panel lets the case owner schedule permanent deletion of the case's substantive contents: scans, prompts, Agent threads, Insights, and Case Context. See Deleting a case's data.
7-day grace period. Confirming the action schedules the deletion for 7 days later. During that window you can cancel from the same screen, or simply re-subscribe: Claira automatically aborts the scheduled deletion if a plan becomes active again before the run date.
Billing and audit records are preserved. Per-case Stripe linkage, the token usage ledger, plan-tier history, and billing emails are kept after the deletion runs so financial reconciliation and SOC 2 audit windows are unaffected.
No path to delete data while a plan is active. The button is visible but disabled until the plan is cancelled, to make the action discoverable for the case owner without enabling accidental data loss for paying customers.

Release 019 -- May 23, 2026

Australian deployment: Claira is now available in Australia.

Highlights

New deployment region: Australia. Claira now runs in two independent regions. Australian customers connect to https://nuix-au.claira.to; Canadian customers continue at https://nuix-ca.claira.to. Each customer's data stays in the region they use: no cross-region replication, no cross-region egress of customer content.
Region pinning enforced at the platform layer. The Australia deployment runs inside a Google Cloud Assured Workloads workload with the Australia Regions and Support control package, which pins every resource to Australian regions (australia-southeast1 and australia-southeast2).
CMEK on Australia. The Australian Cloud SQL database and the media scan staging bucket are encrypted with customer-managed encryption keys whose key material is stored in the Australian region. The Canadian deployment continues to use Google-managed encryption keys.
Pricing in AUD. The Starter / Pro / Business / Enterprise tiers are now available in Australian dollars at the same numeric amounts as the Canadian dollar tiers ($49 / $499 / $4,999 / $29,999 AUD per month).
Updated Privacy and Security page to describe the two-region architecture, the Australia Assured Workloads control package, and alignment with the Australian Privacy Principles and the Notifiable Data Breaches scheme under the Privacy Act 1988.
No change in how you use Claira. The URL you visit determines the region you reach. Nothing else about the experience is different between the two deployments.

Release 018 -- May 23, 2026

Insights: turn bulk-scan results into a single summarised output.

Highlights

New mode: Insights. A new Insights entry in the mode menu lets you crunch the AI results from a completed bulk scan (or from a document selection in any field with values) into a single distilled output (chronology, relevance summary, PII report, investigation narrative) with green-italic citations to each source document inline. See Insights.
Revise without starting over. Each Insight has its own revision history. Type "make it shorter" or "add a paragraph about X" and Claira produces a new revision while preserving the original; you can switch between revisions in the sidebar or roll back any time.
Download as PDF. The Insight viewer downloads the currently selected revision as a PDF with citations rendered green-italic inline and a Source documents table on the trailing pages listing every contributing document by ID and field value.
Fixed model, highest reasoning. Insights always run on Gemini 2.5 Pro at the highest reasoning level: there's no model picker on this path so output quality stays consistent across users.
Coming next. A hidden audit pass that double-checks every cited document against its source text is planned for a future revision, along with a one-click "Generate Insight" button on the bulk-task history detail view and intent-first workflows that propose the bulk task to produce a given Insight.

Release 017 -- May 22, 2026

Consistent "Media scan" vocabulary everywhere.

Highlights

One name for raw-file analysis. Across the action UI, settings, error messages, pricing tables, billing/usage labels, and the documentation, the umbrella term for sending the document's source file (image, audio, or video) to a multimodal model is now Media scan. The specific modalities are Text scan, Image scan, Audio scan, and Video scan. Legacy phrasing like "raw file review" or "Native PDF/image file" has been retired from user-facing surfaces. See Media scans.
Docs URL changed. The Media scans page has moved from /docs/en/ai-review/images to /docs/en/ai-review/media-scans. The old URL redirects automatically; update any saved links at your convenience.
Pricing table is consistent. Bulk and Multi-Code rows in Plans and Billing now use a single "X scan" pattern instead of mixing "image review" / "image scan".
No change to billing, plan gates, or behavior. Token costs, plan tiers, and which modes require Pro are unchanged. Only the names you see have been standardized.

Release 016 -- May 22, 2026

Plan upgrade emails are now friendlier and numbers-first for occasional Claira users.

Highlights

Warmer plan-upgrade email. When a case's plan is upgraded, every case recipient now receives a more welcoming operational email. It reminds the reader what Claira is, frames the change as "you or someone on your team upgraded the plan", and leads with the new monthly token ceiling and bulk-scan limit. Newly-unlocked capabilities (Case Context and Multi-Code coding, when they apply) are listed as a short tail; otherwise the email is a clean numbers-only message. The previous tier is intentionally not shown. See Plans and billing: what your team receives by email.

Release 015 -- May 22, 2026

Single scans are free, and the Usage dashboard now breaks down tokens by bulk type, media, and user.

Highlights

Single scans are now free on every plan. Text, image, audio, and video single scans, including single-document Multi-Code, no longer deduct from the reservoir. The new "Single scans this month" card on the Usage dashboard tracks how many you've run. See Usage Analytics.
New stacked bar chart on Usage. The pie chart has been replaced with a stacked bar that splits tokens between Bulk one-field and Bulk multi-code, each stack broken out by media type (Text, Image, Audio, Video). See Tokens by bulk type and media.
Per-user consumption chart. When more than one user has consumed tokens in the selected month, the Usage dashboard now shows a horizontal bar chart of tokens per user. See Tokens consumed per user.
Planned vs overage pie. A small pie at the bottom of the Usage dashboard appears only when overage tokens exist for the selected month, showing the split between planned and overage consumption. See Planned vs overage tokens.

Release 014 -- May 22, 2026

Bulk scan confirmation now always shows the destination field(s), and bulk-scan size limits follow your plan.

Highlights

Consistent destination field display. No matter which Start Bulk Scan button you use (Prompt Lab, Bulk Scanning toolbar, or Multi-Code), the confirmation popup now shows the destination field in the same place and the same style. For Multi-Code runs the popup displays a Multi-code scan summary with the number of fields that will be written to (it does not enumerate each field). See Bulk Scan.
Bulk-scan document limit now matches your plan. Previously the Claira UI rejected any selection over 25,000 documents regardless of plan. Selections are now compared against your plan's bulk-scan limit: 1,000 on Starter, 10,000 on Pro, 100,000 on Business, and unlimited on Enterprise. See Plans and billing for each tier's cap and Bulk Scan for how the limit is enforced when you start a scan.

Release 013 -- May 20, 2026

New Smart model tier; OSS model removed from the picker.

Highlights

Claira CA: Smart is now available in the Model picker. It runs in Canada under the same residency guarantees as Fast and is tuned for complex, ambiguous, or borderline documents where deeper reasoning helps. Slower per document and more tokens per scan than Fast. Use it on the calls that justify the cost. See Model selection.
Claira Default (CA) has been renamed to Claira CA: Fast (Default). Same model, same behavior; the new name makes the Fast/Smart pair easier to talk about. It remains the default for every case.
OpenAI GPT oss 120b removed from the picker. The model is no longer offered to new selections. Cases that had it selected will now show Fast instead; bulk tasks already in flight that referenced the model continue to run uninterrupted.
Bring Your Own Model is unchanged and still available from the picker.

Release 012 -- May 19, 2026

Media scan article now lists approximate max length per document.

Highlights

Per-mode length guidance. Media scans (image, audio, video) now indicates the approximate maximum length Claira can process in a single document for each mode: ~900 pages for image (PDFs), ~8 hours for audio, and ~40 minutes for video. Use these as a planning reference when deciding whether a long recording or a very large PDF needs to be split before scanning.

Release 011 -- May 15, 2026

New admin article: opt-in controls for sensitive cases.

Highlights

New documentation page. Opt-in controls for sensitive cases explains how Claira's opt-in design at both the case level and user-group level lets administrators keep restricted matters out of AI processing entirely. The page covers the portal-admin and case-admin roles, why leaving Claira off is a hard control, and a recommended policy for organizations that handle a mix of confidentiality levels (for example, government-protected data or regulated customer records).

Release 010 -- May 14, 2026

Single scans are free on every plan; bulk scans no longer charge for failed documents.

Highlights

Single document scans never use tokens. Every single-document workflow (text, image, PDF, audio, video, raw file, multi-step review) is now free on every plan, including Starter. The media plan-tier gate is gone: any plan can scan any media type one document at a time. See Single review, Media scans, and the updated token cost table in Plans and Billing.
Agent Mode (the lab) is free. Messages in a chat thread are single-document interactions and no longer consume a token per message. Use the lab to iterate without watching the meter.
Bulk scans only charge for documents the AI actually scans. When a bulk task ends, Claira automatically refunds tokens for documents that failed, were skipped, or never ran (pending in a cancelled task). Only completed documents stay charged. The refund posts to your billing ledger as a refund entry as soon as the task reaches a terminal state. See Bulk scan: token usage.
Usage reporting reflects the change. The Usage pie chart, the Billing Report PDF, and the in-app category breakdown no longer include single-scan or Agent Mode tokens: those rows are zero by design. Active categories are Bulk Scan, Multi-Code, and Raw-File Review. See Usage and Billing.

What still consumes tokens: bulk scans, Multi-Code single-document runs (1+ tokens per active field), and Multi-Code in bulk. Pricing for those is unchanged.

Release 009 -- May 14, 2026

Knowledge base navigation fixes.

Highlights

New Prompt templates landing page. Hitting the templates section now opens a directory of every available template grouped by category (general, privilege, relevance, objective coding, and investigations) instead of returning a not-found error. See Prompt templates.
Fewer broken links from older bookmarks. Older URLs from before the May 1 navigation restructure (legacy section names like prompting, emails, analytics, use-cases, and renamed pages such as mode-chat-bot) now redirect to their new locations instead of 404ing. Bookmarks, search results, and inbound links from third-party sites should resolve again.
Chronology Summary prompt: cleaner output format. The standard chronology prompt now instructs the model to start each output with YYYY/MM/DD instead of Date YYYY/MM/DD. This makes exported chronologies sort and parse more cleanly, especially when downstream tools split on the first colon. See Chronology Summary template and Chronologies workflow.

Release 008 -- May 2, 2026

Redesigned PDF reports.

Highlights

Bulk Task History Report: fully redesigned. Each task is a self-contained card with the media type badge, success rate, full prompt text in a monospace block, and the customer-facing model name (e.g. "Claira Default (CA)") instead of the raw model id. Clean horizontal dividers between tasks. Renamed from the previous "Coding History Report".
Billing Report: expanded. Now includes a plan section (name + description), a horizontal stacked-bar chart showing tokens by category, and a new all-time scope with a per-cycle breakdown table. The Plan column reflects the highest tier the case held during each cycle, so a mid-month upgrade is captured. Token-factor counts only, with no dollar amounts.
Unified branding: all three downloadable PDFs (Bulk Task History, Billing, Case Context) now render the green-dot + "claira" pill logo from the app header instead of the old raster.
See the updated Exports, Usage, Billing, Bill Claira to your clients, and Bulk scan pages.

Release 007 -- April 29, 2026

Audio and video scans, plus a four-way "Scan as" dropdown.

Highlights

Media scans (new)
- The Image scan checkbox in Single Review, Bulk Scan, and Multi-Code is replaced by a Scan as dropdown with four options: Text (default, 1 token/doc), Image (5 tokens/doc), Audio (10 tokens/doc), Video (20 tokens/doc).
- All non-text modes route the document's actual source file to a multimodal model. Accepted formats: PDF, PNG, JPEG, WebP, GIF, BMP, HEIC, HEIF for image; MP3, WAV, AAC, OGG, FLAC, M4A, AIFF for audio; MP4, MOV, WebM, AVI, MKV, MPEG, 3GP, FLV, WMV for video.
- Audio and video scans require Pro plan or higher, alongside the existing image scan entitlement. Starter users see an upgrade prompt when picking any non-text option.
- See the updated Media scans page and the token cost table in Plans and Billing.
Smarter "missing text" errors
- When a Text scan fails because the document has no extracted text, Claira now inspects the document's source files. If it sees an image, audio, or video file, the error message names the right scan mode ("This looks like audio content. Try Audio scan") instead of the generic "switch source" message. Same suggestion is written into the field for skipped bulk-scan documents.

Release 006 -- May 1, 2026

Prompt Lab naming, richer Agent Mode and templates, and clearer Nuix guidance.

Highlights

Product and UI updates
- Renamed Classic mode to Prompt Lab for the standard single-field workflow; refreshed header layout and in-product search. See Single review and Model selection.
Prompt Lab output
- The AI Response area can show markdown-style formatting in Claira; when you apply output to a Nuix Discover field, the field type may store a plainer version. See Single review.
Templates and workflows
- New and updated investigation-oriented templates (including PII surfacing). Browse PII identification, the Comprehensive PII surface forms template, and the full Workflows collection.
Agent Mode
- Clearer responses, in-app links to help topics, and refined navigation (including a compact sidebar tab when collapsed). Starter plans now include up to 10 saved Agent Mode threads per user per case. Empty-state starter chips now focus on capabilities, explaining the open document, and bulk scan history; the What can Claira do? chip returns a fixed help-desk summary (no model call) with a link to What Claira can do. See Agent Mode, Agent Mode limitations, and Plans and Billing.
Using Nuix Discover
- New guide for common Nuix tasks (workspaces, search, coding, productions) that Claira does not run for you, with pointers to official Nuix help. See Common Nuix Discover tasks.
Reliability
- Improvements to prompt logging and Nuix-related downloads for clearer records and fewer edge-case failures.

Release 005 -- April 14, 2026

Claira relaunch, smarter workflows, and safer billing.

Claira has received a major wave of updates over the last two weeks, with faster review workflows, stronger billing controls, and a new case-scoped agent mode.

Highlights

Product and UI updates
- Refreshed Claira branding, improved scan responsiveness, and polished prompt template and review experiences for day-to-day users.
Core functionality updates
- Expanded single and bulk review workflow controls, improved task handling and messaging, and launched support documentation with richer in-app help links.
Agent mode (new)
- Added case-scoped agent mode for follow-up analysis in case context, with plan-based availability and updated guidance in Agent Mode, Bulk Scan, and Plans and Billing.
Security, billing, and reliability updates
- Improved Stripe-backed activation safeguards, clarified token billing behavior, and hardened deployment/authentication flows for more reliable operations.

Previously released under another product name; this release is the Claira relaunch.

Release 004 -- February 21, 2026

Case context approval workflows, better task management, and clearer error messages.

Case context approval workflows. Case context submissions now go through an approval step before taking effect, giving administrators control over what contextual information is active in the case.
Editable bulk task titles. You can now rename bulk scan tasks to something meaningful -- like the review phase or prompt version -- making it easier to find specific tasks in the history.
Improved error messages. Error messages across the application have been rewritten to be clearer about what went wrong and what you can do about it. See the Error Codes reference for details.

Release 003 -- February 17, 2026

Introducing Case Context, billing reports, and updated pricing.

Case Context (new). You can now provide background information about your case -- key parties, date ranges, legal issues -- so the AI has the context it needs to produce more accurate results. See Case Context Overview to get started.
Billing Report. A new billing report is available that shows token usage, costs, and activity broken down by case and time period.
Case and organization details visible. Case name, organization, and activation status are now displayed clearly within the Claira interface.
Updated pricing. Plan tiers and pricing have been updated. Check the Plans and Billing page for current details.
Improved bulk task status. The bulk scan task list now shows more detailed status information, making it easier to track tasks in progress.

Release 002 -- December 29, 2025

Connect Fields for everyone, and larger bulk scans.

Connect Fields for all users (new). The field connection feature, previously limited to administrators, is now available to all Claira users. Connect Nuix Discover fields directly from the Claira settings panel.
Bulk scan capacity expanded to 25,000 documents. A single bulk scan task can now process up to 25,000 documents, up from the previous limit.

Release 001 -- December 17, 2025

Bulk Scan Task History, user filtering, and expanded prompt library.

Bulk Scan Task History with resume (new). All your bulk scan tasks are now saved in a history panel. If a task is interrupted, you can resume it from where it left off instead of starting over.
Filter task history by user. In shared cases, you can filter the task history to see only your own tasks or tasks from specific team members.
Expanded prompt library. New built-in prompt templates have been added for common review tasks: People extraction, Translations, and Chronology summaries. Browse them in the Prompt area's template picker.

Need help? Contact us at support@claira.to.

Changelog

ChangelogCopy Anchor Link

Release 061 -- October 22, 2026Copy Anchor Link

Relevance Review workflow articleCopy Anchor Link

Help article corrections: skipped documents, error codes, Scan as defaultCopy Anchor Link

Field write errors say what to doCopy Anchor Link

Refunds return to the tokens that paid, and a clearer monthly tokens emailCopy Anchor Link

Field names Claira proposes start with "🟢 Claira - "Copy Anchor Link

The billing ledger shows overage refunds and tokens credited by ClairaCopy Anchor Link

Insight template articles follow the Quickstart dialogCopy Anchor Link

Relevance Review: record the result; plans and billing: refunds at renewalCopy Anchor Link

Release 060 -- October 9, 2026Copy Anchor Link

The open document's chip says Text on a Claira Private caseCopy Anchor Link

A document's file that is too large for a scan in your browserCopy Anchor Link

Giving a teammate access to the Admin PortalCopy Anchor Link

A file the AI model cannot read no longer fails the scanCopy Anchor Link

Very long prompts are kept and open againCopy Anchor Link

A prompt too long for the Prompt LogCopy Anchor Link

Release 059 -- October 7, 2026Copy Anchor Link

Agent Mode knows about the early case assessment as soon as Claira opensCopy Anchor Link

The monthly token email describes the reset correctlyCopy Anchor Link

Pricing wording brought up to date across Claira, the Admin Portal and the help centreCopy Anchor Link

Usage and billing help, and the Buy tokens dialog, describe the monthly resetCopy Anchor Link

A deleted Agent chat is permanently erased after 7 daysCopy Anchor Link

Converting a prompt to Multi-Code works in a Claira Private caseCopy Anchor Link

Claira Private no longer talks about tokens when a scan failsCopy Anchor Link

A Multi-Code answer is shown under the name of the field it is written toCopy Anchor Link

Release 058 -- October 4, 2026Copy Anchor Link

Partner discount set by ClairaCopy Anchor Link

Invoices dated the last day of their billing periodCopy Anchor Link

Reuse documents across Case ContextsCopy Anchor Link

Bulk scans linked to their prompt versionCopy Anchor Link

Bulk task results in one lineCopy Anchor Link

Comments on bulk tasksCopy Anchor Link

Every past invoice in the Admin PortalCopy Anchor Link

Early case assessment PDF looks like the screenCopy Anchor Link

Reuse your Discover API token across casesCopy Anchor Link

A detailed Case Context, carried in full into promptsCopy Anchor Link

Gemini 2.5 models retiredCopy Anchor Link

Large bulk scans start quicklyCopy Anchor Link

Release 057 -- September 29, 2026Copy Anchor Link

Help desk language switcherCopy Anchor Link

Case Context reliabilityCopy Anchor Link

Germany opens in GermanCopy Anchor Link

When a model declines a document, Claira now says soCopy Anchor Link

Export your whole prompt history for the recordCopy Anchor Link

Create a Case Context without typing a nameCopy Anchor Link

Gemini Flash 3.5 comes to AustraliaCopy Anchor Link

Insight reports link their own citations, and hovering a Document ID shows what it saidCopy Anchor Link

PII prompt works outside North America, plus a GDPR Article 9 templateCopy Anchor Link

Chat organization: rename, search, archive, and ordered pinsCopy Anchor Link

Claira can see your InsightsCopy Anchor Link

Move a running scan to the backgroundCopy Anchor Link

Usage figures no longer count the tokens cleared when a case is deactivatedCopy Anchor Link

Applied revisions are kept in Agent ModeCopy Anchor Link

Agent Mode reads the document itselfCopy Anchor Link

Case Context PDFs show approval status and who approvedCopy Anchor Link

Running low on tokens, and the monthly usage emailCopy Anchor Link

Attach documents to an agent chatCopy Anchor Link

Claira remembers the field you are working inCopy Anchor Link

Every exported report now carries page numbers and a footerCopy Anchor Link

Bulk scanning says what you have selected, and past tasks are easier to findCopy Anchor Link

Insights: a Quickstart for templates, the prompt beside the answer, and citations Claira stands behindCopy Anchor Link

Field Migration is now four guided stepsCopy Anchor Link

Prompt Lab: a quieter history list, prompt details worth opening, and a PDF that leads with the final wordingCopy Anchor Link

An Insight now says how many documents it readCopy Anchor Link

Usage calls the first bulk type "Bulk single", names your plan, and fits a narrow panelCopy Anchor Link

Agent Mode cards say where they stand, and the chat list filters like the rest of ClairaCopy Anchor Link

Multi-Code: the card reads top to bottom, and every control says what it doesCopy Anchor Link

Your prompt and your Multi-Code setup save as you typeCopy Anchor Link

Case Context: decisions explained in Claira, template tiles, and one way out of a contextCopy Anchor Link

Settings: two groups you can fold away, model settings that save themselves, and one-click log fieldsCopy Anchor Link

Insights are freeCopy Anchor Link

Prompt History counts scansCopy Anchor Link

Set a logging field from Agent ModeCopy Anchor Link

Case Context saves as you type, captures more, and can be revised with the AgentCopy Anchor Link

Reporting a problemCopy Anchor Link

New Case Context template for arbitrationCopy Anchor Link

New help article: how we choose AI providersCopy Anchor Link

Privacy and Security now covers Germany and states the AI model provider policiesCopy Anchor Link

Changelog

Release 061 -- October 22, 2026

Relevance Review workflow article

Help article corrections: skipped documents, error codes, Scan as default

Field write errors say what to do

Refunds return to the tokens that paid, and a clearer monthly tokens email

Field names Claira proposes start with "🟢 Claira - "

The billing ledger shows overage refunds and tokens credited by Claira

Insight template articles follow the Quickstart dialog

Relevance Review: record the result; plans and billing: refunds at renewal

Release 060 -- October 9, 2026

The open document's chip says Text on a Claira Private case

A document's file that is too large for a scan in your browser

Giving a teammate access to the Admin Portal

A file the AI model cannot read no longer fails the scan

Very long prompts are kept and open again

A prompt too long for the Prompt Log

Release 059 -- October 7, 2026

Agent Mode knows about the early case assessment as soon as Claira opens

The monthly token email describes the reset correctly

Pricing wording brought up to date across Claira, the Admin Portal and the help centre

Usage and billing help, and the Buy tokens dialog, describe the monthly reset

A deleted Agent chat is permanently erased after 7 days

Converting a prompt to Multi-Code works in a Claira Private case

Claira Private no longer talks about tokens when a scan fails

A Multi-Code answer is shown under the name of the field it is written to

Release 058 -- October 4, 2026

Partner discount set by Claira

Invoices dated the last day of their billing period

Reuse documents across Case Contexts

Bulk scans linked to their prompt version

Bulk task results in one line

Comments on bulk tasks

Every past invoice in the Admin Portal

Early case assessment PDF looks like the screen

Reuse your Discover API token across cases

A detailed Case Context, carried in full into prompts

Gemini 2.5 models retired

Large bulk scans start quickly

Release 057 -- September 29, 2026

Help desk language switcher

Case Context reliability

Germany opens in German

When a model declines a document, Claira now says so

Export your whole prompt history for the record

Create a Case Context without typing a name

Gemini Flash 3.5 comes to Australia

Insight reports link their own citations, and hovering a Document ID shows what it said

PII prompt works outside North America, plus a GDPR Article 9 template

Chat organization: rename, search, archive, and ordered pins

Claira can see your Insights

Move a running scan to the background

Usage figures no longer count the tokens cleared when a case is deactivated

Applied revisions are kept in Agent Mode

Agent Mode reads the document itself

Case Context PDFs show approval status and who approved

Running low on tokens, and the monthly usage email

Attach documents to an agent chat

Claira remembers the field you are working in

Every exported report now carries page numbers and a footer

Bulk scanning says what you have selected, and past tasks are easier to find

Insights: a Quickstart for templates, the prompt beside the answer, and citations Claira stands behind

Field Migration is now four guided steps

Prompt Lab: a quieter history list, prompt details worth opening, and a PDF that leads with the final wording

An Insight now says how many documents it read

Usage calls the first bulk type "Bulk single", names your plan, and fits a narrow panel

Agent Mode cards say where they stand, and the chat list filters like the rest of Claira

Multi-Code: the card reads top to bottom, and every control says what it does

Your prompt and your Multi-Code setup save as you type

Case Context: decisions explained in Claira, template tiles, and one way out of a context

Settings: two groups you can fold away, model settings that save themselves, and one-click log fields

Insights are free

Prompt History counts scans

Set a logging field from Agent Mode

Case Context saves as you type, captures more, and can be revised with the Agent

Reporting a problem

New Case Context template for arbitration

New help article: how we choose AI providers

Privacy and Security now covers Germany and states the AI model provider policies

One name for GPT-OSS 120B