Prompt HistoryCopy Anchor Link

Every prompt used or saved on your case is kept in Prompt History, shared with everyone working in the case. Open it with the View history button in Prompt Lab, in Multi-Code, or in the Bulk Scanning section, then restore any entry back into your prompt box.

Because teams refine prompts by re-running small edits, a case's history quickly fills with near-identical entries. Prompt History opens in a Grouped view that stacks those variations into version chains, so you can see how a prompt evolved and which iteration the team settled on. One row above the list holds everything you can do to it: the search box, the view switch, the filters, and the spreadsheet export.

The Grouped viewCopy Anchor Link

The Grouped view is the default. Claira automatically detects prompts that are variations of the same instructions (including edits made by different teammates) and collapses them into a single card:

• The card leads with the prompt itself, in ordinary reading text, trimmed to two lines. Widen the Claira pane and you see more of it.
• A pin and a green check sit in front of the prompt when it is pinned or marked final, and a small wand when one of Claira's AI tools wrote it (see Prompts written by AI). Hover any of them to read what it means.
• Under the prompt, one quiet line: the authors, a versions chip counting the distinct versions in the chain, a used chip when the same wording has been used in more than one scan (see How uses are counted), and the number of comments. A Multi-Code configuration shows its badge, its number of fields and its destination fields in the same line.
• There are no dates on a card. They are in the prompt's details, where there is room to say which date it is.
• Restore (or Insert Template, depending on where you opened the history from) is the card's only button. A Multi-Code configuration has Load into Multi-Code instead: see Multi-Code history.
• Prompts with no variations appear as ordinary single rows, exactly like the chronological view.

Click anywhere on a card to open that prompt's details. The card is reachable from the keyboard too: tab to it and press Enter. The chevron on the right, and only the chevron, expands the version chain, so opening a prompt and unfolding its history are no longer the same gesture.

Inside an expanded chain:

• The version marked final comes first, then the rest newest first. That puts the wording your team settled on, and the wording most recently tried, where you look first.
• Every version keeps the number it was given in the chain and still compares itself against the version it was edited from, whatever order the list is drawn in. The third row down can read "Version 2" and compare with version 1.
• A version that was scanned several times without changes appears once, with a Used N× count instead of duplicate rows.
• From version 2 onward, the prompt text is rendered as a comparison with the previous version: added words are highlighted in green, removed words are struck through in red, and unchanged text is dimmed. A +N −M words chip summarizes the size of each step.
• A version row carries its numbered dot, the final marker and, when an AI tool wrote that version, the wand, and nothing else. Click the row to open that version's details, where the pin, the mark and the download live.

Grouping is automatic and based on text similarity: there is nothing to set up, and it applies to your existing history retroactively. A Multi-Code configuration is only ever grouped with other configurations, never with a Prompt Lab prompt, and two configurations that send the same prompts to different fields are two versions.

How uses are countedCopy Anchor Link

The Used N× count covers scans only:

• Every single-document scan, bulk scan and Multi-Code scan that used the wording counts, including repeats of the same scan. A bulk scan you approve on an Agent Mode card counts too, under your name.
• Hover the count to see the split: Single scans, Bulk scans and, for older prompts, Earlier uses.
• Saving a prompt with Save Prompt, inserting one from history, a template, the Prompt Generator or Agent Mode, or applying a Revise with Agent suggestion, keeps it in history but does not count as a use. Resuming a bulk task, or sending a running scan to the background, does not count the scan again.
• Entries recorded before Claira told scans apart count as one earlier use each, and those can include saves and inserts. An older prompt that was saved or inserted many times can therefore show a lower count than it used to.

Prompts written by AICopy Anchor Link

When one of Claira's AI tools wrote a prompt, a small wand sits in front of it on its card and on its version row. Hover the wand to see which tool:

• Suggested by Revise with Agent: a suggestion you applied from the Revise with Agent popup under a response. It is saved to Prompt History the moment you apply it, so the history shows every revision the AI proposed and you accepted, even before you scan with it. See Rate responses and revise with Agent.
• Written by the Agent: a prompt you inserted from an Agent Mode chat, or the prompt of a bulk scan you approved on an Agent card. When the card runs the wording that was already in your prompt box, there is no wand: that prompt is yours. Nor is there one when you picked the card's prompt from a template with Quickstart, even if you then edited it: the Agent did not write it. When the Prompt Generator wrote the card's prompt, the wand names the Prompt Generator instead.
• Written by the Prompt Generator: a prompt the Prompt Generator drafted for you. That includes the prompt of a bulk scan you approved on an Agent card when the Prompt Generator wrote it: a prompt you inserted from the Prompt Generator with Quickstart on the Bulk scan proposal card, including a template it filled in with Use with context, or one the Agent had the Prompt Generator draft from your goal. Editing it on the card keeps this wand.

Adding AI text to a prompt that already had text (Append rather than Replace) also gets the wand, because the AI wrote part of that revision.

When the Prompt Generator built a prompt from a Case Context and you inserted it with Replace, prompt details also say which context and which version of it, for example Written by the Prompt Generator · Built from Main litigation · version of October 1, 2026. If the context has changed since, a note says so and suggests generating the prompt again to use the latest version. If the context has been deleted, the line says that instead.

A few things to know:

• The wand follows the wording. If the same wording was already in the history before an AI tool produced it (you typed it and saved or scanned it first), it has no wand. Restoring an AI-written prompt keeps the wand, in both views.
• Templates get no wand. A template is not written for you by an AI.
• No wand does not mean a person wrote it, only that Claira has no record that an AI did. Prompts saved before this feature have no wand, whoever wrote them.
• Prompt details say the same in words. Under the prompt, a line with the wand names the tool. For a Revise with Agent revision you applied yourself, Open Agent chat beside it opens the chat where the revision was recorded. Teammates see the line but not the link, because that chat is yours.

Marking a final versionCopy Anchor Link

When a chain has a winner (the wording your team agreed to run with), mark it as the final version:

• In an expanded chain, click the green check on that version, or open the version's details and click the check in its header. Hover either one and Claira explains what the mark means and that only one version can carry it.
• The final version gets a green Final badge, moves to the top of the expanded chain, becomes what the collapsed card previews, and is what the card's Restore button inserts.
• The mark is shared: everyone on the case sees the same final version, and anyone on the case can set it. Marking a different version simply moves the mark; the prompt's details show who marked it and when.
• To clear it, click the check again. Claira asks you to confirm, since this changes what teammates see.

The Chronological viewCopy Anchor Link

Prefer the flat list? Open the view switch in the row above the list and choose Chronological. This is the classic view: every saved entry in date order, with pinned prompts floating to the top. Final versions keep their green check here too, and cards work the same way: click to open the details, Restore to use it. Your choice of view is remembered per case and follows you across devices.

Pins, comments, and searchCopy Anchor Link

Pins and comments work in both views. Pinning a version (for yourself, or for the whole case) floats its group to the top of the Grouped view; comment counts add up across a group's versions.

Agent Mode can read a prompt's comments and, when you ask, add one in your name; a comment the Agent adds is marked Added by the Agent. Deleting a comment is recorded 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.

The search box matches every version in a chain, so a group stays whole even when only one of its versions matches. Beside it, the funnel icon opens the filters:

• Users: everyone, you, or any combination of the people who wrote prompts on this case.
• Date: any time, or the last 7, 30 or 90 days.
• Status: pinned, marked final, or both.
• Multi-Code history: only Multi-Code configurations and the prompts inserted into Multi-Code. See Multi-Code history.

A small number on the funnel says how many of the four you have applied, and Clear filters at the foot of the panel puts them all back. Filters are not remembered: every time you open Prompt History it starts afresh, with Multi-Code history on when you open it from Multi-Code and off everywhere else.

Prompt detailsCopy Anchor Link

Opening a prompt gives you one version in full:

• The header names it ("Version 3 of 5") and carries the Final badge when it has one, with the pin, the final marker and Download PDF beside it.
• Arrows next to the version name step to the previous and next version of the chain, so you can read a prompt's evolution without going back to the list.
• The prompt text comes first, in full, and under it one quiet line: the authors, when it was first and last used, how many single and bulk scans used it (or Not used in a scan yet), and who marked it final. When an AI tool wrote the prompt, a line with the wand above it says which one (see Prompts written by AI).
• Changes vs previous version is a section you open when you want the comparison and leave closed when you do not. Claira remembers which way you left it.
• Bulk scans with this version lists every bulk scan in the case that ran this wording. See Bulk scans that ran a version.
• Comments follow: read the thread, add one, or delete your own.
• At the foot, Restore (or Insert Template) applies the prompt, named exactly as it is on the card. A Multi-Code configuration offers Load into Multi-Code instead.

The case name is no longer shown. Prompt History only ever lists the case you are already in.

Bulk scans that ran a versionCopy Anchor Link

Every bulk scan is linked to the version of the prompt it ran. You can see which scans used a wording, and check that a scan used the wording your team settled on.

• In a prompt's details, Bulk scans with this version lists every bulk scan in the case that ran exactly this wording, newest first, whoever started it. Each one shows its log reference (such as SC014, the value written to the Log Field), its name, its status, who ran it, when, and how many documents it coded. For a Multi-Code configuration, the list holds the bulk scans that ran that configuration.
• Select a scan to open its task details. Close them to come back to the prompt.
• Differences in spaces and line breaks do not count, just as they never split a version. Any other change to the wording is a different version, with its own list.
• The list includes bulk scans run before this link existed and bulk scans approved on an Agent Mode card. It can therefore hold more scans than the Bulk scans count above it, which started when Claira began telling single and bulk scans apart (older uses are counted as Earlier uses).
• When more than 100 bulk scans ran a version, the 100 most recent are listed, and a line under the list gives the total.

It works the other way too: a bulk scan's task details name the version it ran, and say when it was not the version marked final. See Which prompt version a task ran.

Downloading a prompt as PDFCopy Anchor Link

Open a prompt's details and click Download PDF: the document to attach when you send a prompt out for approval, or to file the wording that was actually run. The download is in the details rather than on every row, so the list stays a list. Prompts with more than one version offer a choice:

• This version only: a short document with the prompt's full text, who wrote it and when, its place in the chain ("Version 4 of 6"), how many scans used it (with the single and bulk split) when it was used more than once, the final mark if it has one, and the comments left on that version.
• Full version history: the wording your case settled on, first and in full, with its comments directly beneath it, and that section has the first page to itself. That is the version marked final, or the current version when nothing is marked. Version history starts the next page: every version, newest first, with what changed from the version before it (added words in bold green, removed words struck through in red, with a +N −M words count), its complete text, and its own comments underneath. The featured version is not printed twice: it keeps its numbered place in the list with a one-line pointer back to the top.

A prompt with no variations downloads straight away, without the choice.

Both documents are titled Prompt export and carry the Claira header, the case name, and who exported it and when. Every page ends with a line naming Claira, the report, the case, and the date you exported it, with Page 2 of 9 opposite on the right, and from the second page onward a small running header repeats the title at the top of the page. The prompt's opening words name the file rather than the document: the prompt itself is printed in full on the first page.

Every download is recorded for your organization's compliance records: who exported the prompt, which version, and when. The record is written before the file is handed to you, so on the rare occasion it cannot be saved, the download is not created and Claira asks you to try again. The record holds no prompt or comment text, only counts, including how many characters, if any, had to be replaced (see below). The filename follows the pattern Claira-Prompt-<opening words>-v<N>-<YYYYMMDD>.pdf, or Claira-Prompt-<opening words>-history-<YYYYMMDD>.pdf for the full history.

If the prompt or its comments contain characters the PDF font cannot show (Chinese or Cyrillic text, or mathematical symbols such as ≥), each one appears as ¤ in the document, the last line of the document says how many were replaced, and Claira tells you after the download. Accented letters, the euro sign, œ and « » print normally.

Exporting the whole prompt historyCopy Anchor Link

The PDF above covers one prompt. To take the entire case (every prompt, each of its revisions, and the comments left on them), click the download icon at the end of the row above the list. Its tooltip reads Export history.

This is the export to reach for when you need to show what was tried and when: which wordings were run, how each was refined, who changed what, and why. You choose:

• What to export: everything in the case, or only what your current search and filters are showing. (The scope choice appears only when something is narrowing the list.) A search matches whole chains, so a prompt is never exported with revisions missing from the middle of it. The summary sheet names the search, the users, the date window and the status you had applied, and says Multi-Code history only when that filter was on, so the reader of the file can tell which subset they are holding.
• Include: comments, and the change summary between revisions. Both are on by default.
• Format: Excel (.xlsx) or CSV.

The Excel workbook has three sheets:

• Export summary: the case, who exported it and when, what the export covered, and how many prompts, revisions and comments it contains.
• Prompt history, one row per revision: its prompt number and revision number, the full prompt text, author, first and last use, times used, pin state, the final mark and who set it, how many words were added and removed since the previous revision, and the text that changed. At the end of each row: Kind (Prompt, Multi-Code configuration or Multi-Code prompt), Fields (how many a configuration has), then Single scans, Bulk scans and Earlier uses, which add up to the times used, and last Origin: Revise with Agent, Agent, Prompt Generator or Template, left blank when Claira has no record of who wrote the revision.
• Comments, one row per comment: the prompt ID, prompt and revision numbers, author, comment text, date, whether it was later edited, and whether the Agent added it (Added by the Agent).

The Prompt ID column appears on both sheets, so you can line a comment up with the exact revision it was left on.

CSV is a single flat file instead: the same one-row-per-revision grid, with each revision's comments folded into a final column.

Every date in the export is UTC, written in ISO 8601 format (2026-08-01T10:00:00Z), so the dates sort correctly and read the same way in every region.

Anything the export could not include is stated on the summary sheet rather than left for you to spot: if comments failed to load, for instance, the file says so on its face.

As with the PDF, every export is recorded for your organization's compliance records: who exported the history, in which format, how much of it, and when. The record is written before the file is produced, so if it cannot be saved the export is cancelled and Claira asks you to try again. The record holds no prompt or comment text. The file is named claira-prompt-history-<case>-<YYYYMMDD>.xlsx (or .csv).

Very long promptsCopy Anchor Link

A prompt can be up to 1,000,000 characters long, which leaves room to paste a long report or a full set of instructions into it. A prompt of that length is saved in your prompt box, kept in Prompt History and opened in prompt details like any other.

• Longer than that, a prompt still scans but is not kept. Prompt Lab's save status reads Your changes were not saved, with the reason that the text is too long to save, and scans that use it are not added to Prompt History. Shorten the prompt if you need a record of it.
• When a case holds many very long prompts, Prompt History lists the pinned and the most recently used of them rather than all of them, so that the list still opens quickly. The export of the whole prompt history (see above) includes every one.
• A very long prompt is not written to the Prompt Log. A Prompt Log entry sits on the document itself, and a bulk scan adds one to every document it codes, so its size is limited to about 300,000 characters; a prompt with many line breaks or symbols reaches that limit sooner. When your prompt is over it:
  • In a bulk scan, the Write the prompt to each document's Prompt Log checkbox in the bulk scan confirmation is unavailable and a note says why. The scan runs without the Prompt Log; 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.
  • When you save a single result with Also write the prompt to the Prompt Log ticked (see Recording the prompt in the Prompt Log), the value is saved, nothing is added to the document's Prompt Log, and the message after the save says so.

Multi-Code historyCopy Anchor Link

Each Scan Document and Start bulk scan in Multi-Code records the configuration it ran: the shared instruction and every field row, with its destination field and prompt. Running the same configuration again adds to its count instead of creating a new entry.

• Open Prompt History from Multi-Code and the Multi-Code history filter is already on, so the list shows configurations and Multi-Code prompts. Turn it off in the filters to see every prompt in the case. It is on again the next time you open the history from Multi-Code.
• A configuration card carries a Multi-Code badge with its number of fields, followed by its destination fields. Hover the badge to read the full list.
• Its details show the shared instruction, then each field's name, destination field and prompt.
• Load into Multi-Code, on the card and at the foot of its details, puts the whole configuration back into the Multi-Code card: the shared instruction and every field row, with its prompt and destination field. This is how you rerun an earlier pass, such as last month's relevance review. Opened from Prompt Lab or Bulk Scanning, it also switches you to Multi-Code.
• Loading replaces everything in the Multi-Code card. If the card holds anything, Claira asks first. What you replace stays in Prompt History only if you have already scanned with it.
• Each field keeps its destination field while that field still exists in the case. If it has been deleted, Claira uses the field with exactly the same name; if there is none, the row shows Choose a destination field and is skipped until you pick one. Values already written to documents in Nuix Discover never change.
• Prompts you inserted into the shared instruction or into a Multi-Code field are listed as Multi-Code prompts, and you can restore them as before.
• To start from one of Claira's own Multi-Code setups instead, see Multi-Code templates.

Restoring promptsCopy Anchor Link

Restoring works the way it always has: in Prompt Lab and Bulk Scanning, Restore replaces or appends to your current prompt (you're asked when the box already has text). In Multi-Code, Claira asks whether to insert into the shared instruction, an existing field, or a new field row. See Custom Prompts for guidance on refining what you restore.