Bulk Scan

A bulk scan lets you apply the same prompt and field connection to many documents at once. You can include up to 999,999 documents in a single bulk task, on every plan — see Plans and billing for token capacity by tier. Instead of scanning documents one at a time, you select a set of documents in Nuix Discover's browse list and let Claira work through all of them automatically. This is how you move from prompt testing to production-scale review.

Before you run a bulk scan

You need three things in place before starting:

1. A connected field. In the Claira Settings area, connect the Nuix Discover field where results should be written. If you have not done this yet, see the Fields guide.
2. A saved prompt. Write and save your prompt in the Prompt area. We strongly recommend testing it first with Single Review on a sample of documents. Single review is free on every plan and every media type — bulk scans are the only thing that consume tokens.
3. Selected documents. In Nuix Discover's browse list, select the documents you want to scan. The current selection determines which documents Claira will process.

How to run a bulk scan

1. In Nuix Discover, select the documents you want to scan from the browse list.
2. Open the Claira pane and choose one of these entry points:
   • Go to Bulk Scanning mode.
   • Stay in Prompt Lab mode and use the bulk-start button in the Prompt section.
3. Confirm your prompt and connected field are correct.
4. Review the confirmation details. Claira shows you:
   • The estimated token usage for this run next to your current token balance, at the top of the dialog, with a confirmation of whether the run fits within your balance or may incur additional fees
   • The number of documents in your selection
   • Your current scan input mode
   • The estimated duration for the task
   • The destination field where results will be written. For a Multi-Code run, this row instead shows a Multi-code scan summary with the number of fields that will be written to.
   • On cases where Background Processing is set up, a Run as background processing task checkbox when the run qualifies: Text scans on a Claira default model, for both single-field and Multi-Code runs (for Multi-Code, its own Scan as selector must be on Text).
5. Start the task:
   • In Prompt Lab mode, click (shown when more than one document is selected).
   • In Bulk Scanning mode, click .

Monitoring progress

Once the scan starts, you can watch progress in real time within the Claira pane. The progress indicator shows how many documents have been processed out of the total.

A live results panel streams each completed document as it finishes, newest first, so you can watch results land in real time. To read one in full, hover your cursor over it: the entry expands, and the feed pauses while your cursor is on the list — so the entry you're reading stays in place instead of being pushed down as new results arrive. Move your cursor off the list and the feed resumes, adding everything that arrived while you were reading. For a Multi-Code run, each entry shows the extracted values as Field: value lines — one per destination field — the same way the single-document Multi-Code response is presented.

While the scan runs, the counters above the live results track each outcome separately: processed counts the documents that produced a result — the same ones listed in the live results panel — while skipped counts documents with no readable text or media (or whose text is over the size limit), and failed counts documents that errored. Only documents that produced a result appear in the live results panel; skipped and failed documents are reflected in the counters only.

Adjusting the scan speed

Live bulk scans — the runs that stay in your browser tab — include a Scan speed slider in the progress area, with five positions from Slowest to Fastest. It controls how many documents Claira works on at the same time:

• Slower positions ease the load on the AI service. Use them when you see documents failing with resource-availability errors — at the slowest settings these errors all but disappear.
• Faster positions finish the task sooner, but are more likely to run into resource-availability errors, especially during busy hours. Documents that fail this way are refunded automatically and can be scanned again from the Bulk Tasks list.
• Balanced, the default, deliberately runs a little below Claira's maximum speed — the sweet spot between finishing quickly and keeping resource errors rare. Every new scan starts at Balanced.

You can move the slider at any time while the scan runs. The change takes effect right away for the next documents — documents already being scanned simply finish at their own pace — and the estimated duration updates to match the new speed. The slider applies to live runs only; background processing tasks are paced on the server.

There is no need to interact with the interface while the scan runs — adjusting the speed is entirely optional. Just keep the browser window open and let Claira work through the list.

Monitoring background tasks

When a bulk scan runs as a background processing task, the ongoing-task card in the Bulk Tasks list shows a color-coded progress bar that fills from left to right as documents move through the run: green for results already written back to Nuix Discover, teal for documents whose AI processing is complete and that are waiting to be written back, blue for documents submitted for AI processing, and yellow for documents still having their text extracted. Beneath the bar, a count line for each phase appears only while it is informative — the extraction line disappears once extraction is done, and phases with a zero count are hidden. The numbers update whenever you refresh the Bulk Tasks view — there is nothing to configure, and you can close Nuix Discover at any time; the run continues server-side and Claira emails you when it finishes.

Background runs are built to ride out interruptions on their own: brief service hiccups, restarts, and long waits are absorbed silently and the task simply keeps going. Very long-running scans are continued automatically behind the scenes as well — the task just keeps going and the bar keeps advancing; you never need to restart anything. If you receive a failure email, it means the run genuinely stopped making progress and needs attention — open Bulk Tasks and use Resume to continue it from exactly where it left off.

Background runs support single-field and Multi-Code scans alike. A Multi-Code background task codes every one of your configured fields for each document, exactly as the foreground run would; if some fields could not be written for a document, the task's document log says which ones.

Task names

Claira names each bulk task automatically from your prompt shortly after the run starts. You can change the name at any time:

• Rename it — click the task name in the Bulk Tasks list or at the top of the task details, type the new name, and press Enter (or click away) to save. Press Escape to cancel.
• Ask the AI for a new name — in the task details, click the small refresh icon next to the name (its tooltip reads Regenerate title with AI). Claira rereads your prompt and proposes a fresh name — useful when automatic naming didn't complete or the name doesn't capture what the run was about.

If a scan stops early

A bulk scan runs in your browser tab. If the run is interrupted before every document has an outcome — for example the tab was closed, the connection dropped, or a burst of AI-service errors stopped the workers — the task is marked Incomplete instead of Completed, so unscanned documents are never silently presented as done.

An Incomplete task is easy to finish later:

• Open the Bulk Tasks tab and look for the amber Incomplete badge. The task details show exactly how many documents still need scanning.
• Click Resume Task to scan the remaining documents. Documents that already have results are not re-scanned and are not charged again.
• If you prefer not to finish the run, click Cancel task in the task details. The tokens reserved for the unscanned documents are released back to your balance.

While a task is Incomplete, the tokens reserved for its remaining documents stay on hold so that resuming can never charge you twice. They are settled automatically when the task finishes or when you cancel it.

Cancelling a background task works differently. Clicking Cancel task on a background run first opens a dialog: if documents have already finished AI processing, it tells you roughly how many — "Roughly N documents have finished processing" (or generic wording when the count is not yet known) — and offers three choices:

• Write back results & cancel — Claira saves the completed results to the destination field before stopping the task. You are charged only for the documents actually written; the tokens reserved for everything else are returned to your balance.
• Discard & cancel — the task stops without writing anything further. This is permanent — the completed-but-unwritten results cannot be recovered — and the tokens reserved for the unwritten documents are returned to your balance.
• Keep running — closes the dialog and lets the task continue as if nothing happened.

No email is sent for a cancelled task.

Background runs also protect completed work when something goes wrong while nobody is watching: if a scan hits a failure it cannot recover from, Claira automatically writes back everything that had already completed before marking the task failed, and the notification email reports honest counts of what was written and what was not. Documents that could not be completed get an explanatory ERROR_LOG note in the destination field, just like other per-document errors.

Where results go

Results populate directly into the connected Nuix Discover field while the task is still running. You do not need to export or transfer anything -- the data flows straight into your existing review workflow. Foreground scans write each document's result as it completes, and background scans -- including very large ones -- write results back progressively in batches throughout the run rather than all at once at the end.

You can begin reviewing results in Nuix Discover as soon as documents start completing, even while the rest of the task is still running. Results that have already been written are kept even when a task does not run to completion: if you cancel a background task and choose to write back the completed results, or a run fails partway through, everything that finished stays in the field.

Bulk Task History Report

For an audit-friendly record of every bulk task you've run on a case, click the download icon at the top of the Bulk Tasks tab. The PDF contains one card per task with:

• The task number, name, and media type badge (auto, text, image, audio, or video).
• Status, success rate, and the count of succeeded / failed / skipped documents.
• The full prompt text that was sent to the model.
• The customer-facing model name (e.g. "Claira Default (CA)") and the run configuration (temperature, language, destination field).
• Tokens used.

This is the canonical record of what was run on the case — useful for QC reviews, defensibility, or sharing with a client.

Token usage

Token usage depends on the Scan as dropdown selection. The per-document rate is charged only for documents the AI successfully scans — failed, skipped, and pending (cancelled) documents are refunded automatically when the task reaches a terminal state.

• Auto mode (default): 1 to 20 tokens per successful document, set by the mode Claira picks for that document (text=1, image=5, audio=10, video=20). The task reserves the worst-case rate (20 / doc) up front and refunds the difference as each document resolves.
• Text mode: 1 token per successful document
• Image mode (single field): 5 tokens per successful document
• Audio mode (single field): 10 tokens per successful document
• Video mode (single field): 20 tokens per successful document

When Multi-Code is enabled, the per-document cost is the base media scan rate plus 1 flat token — the same regardless of how many active multi-code fields you configure (1 to 8 fields all cost the same):

• Text + Multi-Code: 2 tokens per document.

• Image + Multi-Code: 6 tokens per document.

• Audio + Multi-Code: 11 tokens per document.

• Video + Multi-Code: 21 tokens per document.

• Claira reserves the full task cost upfront and checks your token balance before starting; a scan will not begin if it would exceed your available tokens.

• When the task finishes, any documents that did not complete successfully are credited back to your reservoir as a refund entry on the billing ledger.

• To check your current balance, look at the token count displayed in the bulk scan confirmation dialog or in the Usage popover.

What counts as a successful document

A bulk task processes every document you selected. At the end of the task each document is in one of these states:

• Completed — the AI returned a response. Charged.
• Failed — the AI returned an error, the model could not parse the file, or retries were exhausted. Refunded.
• Skipped — the document type is unsupported in the selected mode or the content is unreadable. Refunded.
• Pending — the task ended (cancelled, abandoned, or terminated for another reason) before the document was scanned. Refunded.

If the task is Incomplete (the scan stopped early — see If a scan stops early), the not-yet-scanned documents stay Pending and their reserved tokens stay on hold until you resume the task or cancel it.

The success / failed / skipped counts are also visible per task in the Bulk Task History Report.

• Keep one saved configuration per workflow to reduce setup drift.
• Record model, template, and field mapping choices in your matter log for auditability.
• When results vary unexpectedly, rerun the same sample before changing multiple settings at once.

Scan-as dropdown (auto / text / image / audio / video)

Bulk scan uses Auto by default — Claira picks the best mode for each document at scan time, so a mixed production gets the right treatment without you needing to sort it first. Auto, Text, Image, Audio, and Video are all available on every plan, including Starter.

• Auto (default): Claira inspects each document and picks Text / Image / Audio / Video on the fly. Bills at the resolved mode's rate per document (1 / 5 / 10 / 20 tokens). Up to 20 tokens / doc is held at task creation; the difference is refunded as each document resolves to a cheaper mode. See Media scans for the full hierarchy.
• Text: Claira sends Nuix extracted text. 1 token/doc (single-field), or 2 tokens/doc in Multi-Code.
• Image: Claira attaches the source image / PDF file. 5 tokens/doc (single-field), or 6 tokens/doc in Multi-Code.
• Audio: Claira attaches the source audio file. 10 tokens/doc (single-field), or 11 tokens/doc in Multi-Code.
• Video: Claira attaches the source video file. 20 tokens/doc (single-field), or 21 tokens/doc in Multi-Code.

What the model receives by mode

• Text mode: Claira sends your prompt plus Nuix extracted text in the request body.
• Image mode: Claira sends your prompt and attaches the source document. Accepted formats: PDF, PNG, JPEG, WebP, GIF, BMP, HEIC, HEIF.
• Audio mode: Claira sends your prompt and attaches the source audio file. Accepted formats: MP3, WAV, AAC, OGG, FLAC, M4A, AIFF.
• Video mode: Claira sends your prompt and attaches the source video file. Accepted formats: MP4, MOV, WebM, AVI, MKV, MPEG, 3GP, FLV, WMV.

If a file section is unreadable or missing, Claira prompts the model to state that limitation and provide best-effort findings.

If a scan fails because of input mode

When a document cannot be scanned in the selected mode, Claira shows an actionable error. Where possible, Claira inspects the document's source files and tells you which mode would work — for example, a Text scan on an audio-only document writes "this looks like audio content — try Audio scan" into the field.

Common examples:

• Source file unavailable or unsupported MIME for the selected mode: switch the Scan as dropdown to a different mode and rerun.
• Extracted text is missing on a Text scan: switch to the suggested media mode (or Image if no suggestion is available).

Next steps

• New to prompting? Start with the Prompting Overview.
• Need media scan details (image / audio / video accepted formats)? See Media scans.
• Want to understand your results? See Understanding Results.
• Encountering errors? Check Error Codes or Troubleshooting.

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