Claira could not read the document's text or file.

This error occurs when Claira cannot get a document's extracted text or file from Nuix Discover, or gets a file the scan cannot use: one that is too large, in a format the scan does not read, or that the AI model could not read.

Common causes:

• A network interruption between Claira and the Nuix Discover API
• The Nuix Discover API is temporarily unavailable or experiencing issues
• The document was deleted or is no longer accessible in the case
• In a media scan, the document's file is not in a format the scan can read, or its TIFF or JPEG 2000 image could not be prepared for the AI
• In an Image, Audio or Video scan that runs in your browser, the document's file is larger than 23 MB. The note gives the file's size, and the document is skipped and not charged. See File size limits.
• In an Image, Audio, Video or Auto scan that runs in your browser, the AI model could not read the document's file, for example a PDF with no pages, or a damaged or password-protected file. The document is skipped and not charged, and Claira does not try it again, because the same file gives the same result. To scan the document's extracted text instead, run it again with Scan as set to Text.
• In a background Image, Audio, Video or Auto scan, the document's file could not be downloaded, was larger than 50 MB, was not a readable format for the scan type, or was never uploaded before the scan started. The note in the field says which, and the document is not charged either way. Two of those are skipped rather than failed, because the document simply has nothing the scan can read: no file in a format the scan type reads, and a file larger than 50 MB. In Auto, those two skip the document only when it has no extracted text to fall back on. See Running media scans in the background.

How to resolve:

1. Check that your Nuix Discover environment is online and responsive.
2. Verify that the document still exists in the case and has extracted text.
3. If extracted text is missing for this document type, switch the Scan as dropdown to Image, Audio, or Video to match the document's source file. Claira's "missing text" error often names the right mode automatically. See Media scans for the formats accepted by each mode.
4. Retry the scan when the note points to Nuix Discover or the connection: a brief network issue usually clears on a second attempt. A note about the file itself (too large, not a supported format, or a file the AI model could not read) gives the same result every time, so use the next two steps instead of retrying.
5. If the note says the file is too large, or that the AI model could not read it, set Scan as to Text to scan the document's extracted text instead. A file between 23 MB and 50 MB can also be scanned as a background processing task. See File size limits.
6. If the message is about the document's file format or image file, see Image scan says the file is not a supported format.
7. If a connection error persists, confirm with your Nuix administrator that the API is functioning normally.

The document is too large to scan.

Most often, the document's extracted text is larger than the per-document limit (about 3 MB). Claira then marks the document skipped, not failed, and the rest of the scan continues normally. This guard runs in your browser, before the text is sent to the AI, so a single very large document can't interrupt a bulk scan. The one exception is the last cause below, a file sent with a very long prompt in a media scan, which is recorded as failed. Neither is charged.

Common causes:

• A very long document, such as a large spreadsheet export, a long transcript, or a heavily OCR'd file with a lot of recognized text
• A document whose text extraction produced an unusually large amount of content
• In a media scan that runs in your browser, a file close to the 23 MB limit sent with a very long prompt. The note says the file is too large for a scan in your browser, and the document is marked failed and not charged. See File size limits.

How to resolve:

1. Open the affected document in Nuix Discover and use Unitize to split it into smaller parts, then scan those parts.
2. If the document is mostly tables or images, consider a Media scan instead of a text scan.
3. If the note says the file is too large for a scan in your browser, run the scan as a background processing task, which reads files up to 50 MB, or set Scan as to Text to scan the document's extracted text.
4. In every case, nothing was sent to the AI model.

The AI model failed to process the document.

This error occurs when Claira successfully retrieves the document text but the AI model cannot complete its analysis. If the model specifically declined the request because of a safety or recitation filter, the result will be E-SAFETY instead, which is a separate, non-retryable case.

Common causes:

• The AI service is temporarily unavailable or under heavy load
• A malformed or unexpected request was sent to the model
• The AI model did not accept the request for this document. The note says that scanning it again will give the same result, and Claira does not retry it
• The model encountered an internal error during processing

How to resolve:

1. Wait a moment and retry the scan. Temporary service disruptions usually resolve themselves within minutes. The exception is a note saying that scanning the document again will give the same result: do not retry it, and try a different Scan as setting or another model instead, or contact support.
2. During a live bulk scan, if several documents fail this way in a burst, lower the Scan speed slider in the progress area: slower speeds sharply reduce failures caused by the service being under heavy load. See Adjusting the scan speed.
3. If you are running a bulk scan, check whether the error affects all documents or just a few. Isolated failures often indicate a document-specific issue (such as unusually long or malformed text).
4. Try switching to the other available model to see if the issue is model-specific.
5. If the message says the AI model is busy, turn on automatic model routing: Claira then answers with another approved model in your country instead of reporting the error. It does not cover bulk tasks run with Background Processing.
6. If the error continues across multiple documents and retries, contact support.

The AI declined to answer this document because of a safety or recitation filter.

This error means the document content (or the combination of the prompt and the document) triggered the model's content-safety filter or its recitation-prevention guard, and the model returned no response. Unlike a transient E-AI failure, this block is deterministic: re-asking the same document with the same prompt will produce the same result, so Claira does not retry.

Common causes:

• The document contains material the model is policy-prohibited from generating output for (for example, certain categories of explicit, harmful, or sensitive content)
• A long passage in the document closely matches copyrighted training material; the model's recitation guard suppresses the response to avoid reproducing it verbatim
• A combination of the prompt phrasing and the document content trips a borderline safety classifier
• The selected model applies its own content policy to the request and rejects it outright, before writing any response. Filters differ from model to model, so a document one model refuses is often scanned normally by another

How to resolve:

1. Open the affected document and review the content. If the block is correct (the document genuinely contains prohibited material), no automated scan is possible: handle the document manually or exclude it from the task.
2. Switch to a different model in Settings and scan again. Each model enforces its own content policy, and this is the fastest thing to try when the document looks benign. On a single-document scan, the error message carries a Change model shortcut straight to the picker.
3. If the block looks like a false positive, try rephrasing the prompt. For example, switch from "summarize this document" to "list the key facts mentioned in the document": softer framing sometimes avoids the trip.
4. For recitation blocks on long quoted passages, ask the model to summarize rather than reproduce content. Summary prompts already do this by default, but worth verifying for translation or extraction prompts that might ask for verbatim output.
5. If E-SAFETY appears on documents you believe are clearly benign, contact support with the document ID so the team can review the prompt/document combination.

Could not write the result back to the Nuix Discover field.

This error occurs when Claira processes the document successfully but cannot save the result to the connected Nuix Discover field.

The sentence after the code says why the write failed and what to do. The same sentence appears when you save a single answer to a field.

Causes and what to do:

If the same message keeps coming back after you have followed its step, contact support.

An unexpected error occurred.

This is a catch-all error for situations that do not fall into the categories above. It typically indicates an unhandled edge case in the system.

Common causes:

• An unexpected combination of conditions that the system did not anticipate
• A transient infrastructure issue

How to resolve:

1. Retry the scan. Many unexpected errors are transient and resolve on a second attempt.
2. If the error recurs, note the document ID, the time of the error, and any other details visible in the Claira panel.
3. Contact support with those details so the team can investigate.

Your model endpoint is not responding.

Claira could not open a connection to the endpoint, or the endpoint did not answer before the time limit. Claira retries once before reporting the document, and nothing is charged for it. During a bulk scan, an amber banner appears above the live results once several documents in a row fail this way.

Common causes:

• The endpoint is down, restarting, or overloaded
• A DNS, firewall, or certificate problem between Claira and the endpoint (self-signed certificates are not accepted)
• The endpoint's address changed since the connection was saved

How to resolve:

1. In Settings, under Team settings, open the Shared connection section, click Edit the shared connection, and click Test connection: the Reachability check names what is wrong.
2. If the endpoint was overloaded rather than down, lower Concurrent requests or move the Scan speed slider down a notch.
3. Once the endpoint responds again, open Bulk Tasks and click Resume Task on the Incomplete task. Documents that already have results are not scanned again.

Your model endpoint rejected the API key.

The endpoint answered that the key Claira sent is not valid. As soon as this happens, Claira stops sending requests to that endpoint (the rest of the run is not failed document by document), and the connection in Settings shows Needs attention.

Common causes:

• The key was revoked, rotated, or expired on the provider's side
• The key was pasted with extra characters or is missing a prefix the provider requires
• The endpoint now expects a different authentication scheme

How to resolve:

1. In Settings, under Team settings, open the Shared connection section and click Edit the shared connection. Paste a valid key into API key and click Test & save. The connection clears as soon as the new key is verified.
2. Open Bulk Tasks and click Resume Task to finish the interrupted run.

This model accepts text scans only.

The scan asked for an Image, Audio, Video, or Auto scan on a model that only accepts document text: the model of a Claira Private case, or a Claira model that reads text only (see Which models can run a media scan). Nothing was sent to the model and nothing was charged.

Common causes:

• Scan as was set to a media mode on a Claira Private case
• Scan as was set to a media mode while a text-only Claira model was selected
• A saved scan configuration from another case carried a media mode over

How to resolve:

1. Set Scan as to Text and run the scan again.
2. On a standard plan, for documents that genuinely need a media scan, switch the Model picker to a model that reads that media for that run. A Claira Private case scans text only.

This model can't read this document's file format.

The selected model reads images, but this document's file is in a format it does not read: a PDF, a multi-page scan, or an audio or video file. Nothing was sent to the model, nothing is charged, and a bulk scan does not retry the document. It marks the document skipped, with an E-FETCH note naming the file type, when Claira can tell from the file itself that the model does not read it, as with a PDF. It records the document as failed when that only shows once the file has been prepared for the model, as with a multi-page scan.

Common causes:

• Scan as was set to Image with Llama 4 Maverick, Gemma 4 31B or Qwen 3.8 Max selected, and the document's file is a PDF or a multi-page scan

How to resolve:

1. Set Scan as to Auto, which uses the extracted text when the model can't read the file, or to Text, and run the scan again.
2. For documents whose images matter, switch the Model picker to a Gemini model for that run. See Which models can run a media scan.

This model connection does not support Multi-Code.

When the connection was tested, the model did not return well-formed JSON for the Multi-Code check, so Multi-Code is switched off for this connection. Nothing was sent and nothing was charged.

Common causes:

• The model does not follow structured-output instructions reliably
• The endpoint truncates responses before the JSON is complete (a small Max output tokens value)

How to resolve:

1. Run the fields as single-field prompts on your model.
2. To enable Multi-Code on your endpoint, connect a more capable or instruction-tuned model, raise Max output tokens if it was small, and run Test connection again until the Multi-Code check passes; then Test & save.

This model cannot run as a background task.

A bulk scan was started with background processing while a model that only supports live runs was selected. The task was not created and nothing was charged. On a Claira Private case, the code is PRIVATE_BACKGROUND_UNAVAILABLE instead.

Common causes:

• Background processing was chosen in the bulk-scan dialog while a model that runs live only was selected

How to resolve:

1. Start the bulk scan as a live run and keep the browser window open until it finishes.
2. If the run must happen in the background, switch the Model picker to a Claira model first.

The case is set to its shared connection, but no connection is saved.

The scan was meant for the case's shared connection, yet the case has no saved endpoint, so there is nowhere to send it. Nothing was sent and nothing was charged.

Common causes:

• The connection was revoked
• The connection was never completed with Test & save
• A window that stayed open from before the connection was revoked

How to resolve:

1. In Settings, under Team settings, open the Shared connection section and complete the connection with Test & save, or
2. Turn on a local model for your own work.

Shared connections are turned off for this deployment.

Your Claira deployment does not currently allow connecting external model endpoints. The Shared connection panel shows it as unavailable, and scans sent to a shared connection are refused.

How to resolve:

1. Turn on a local model to keep scanning.
2. Contact support@claira.to if your organization needs shared connections enabled.

Too many connection tests for this case in the last hour.

Test connection can be run 10 times an hour for each case; the eleventh is refused. Nothing is wrong with the saved connection: it keeps scanning normally. Only running the test again is refused.

Common causes:

• Working through a configuration problem by testing over and over
• Several people in the same case testing the same connection at the same time

How to resolve:

1. Wait and run the test again. The limit is a rolling hour, so it clears as the earlier tests age out.
2. In the meantime, read the check rows from the last test rather than re-running it: they already name what failed.
3. A connection that is already saved does not need re-testing to keep scanning. Test it again only when something about the endpoint has actually changed.

The bridge window is closed or not responding.

Claira reaches your local model through a small separate window, the bridge, and that window is where your model is actually contacted. It has to stay open while Claira is working. When it closes, Claira stops and offers to reopen it rather than failing document after document.

Common causes:

• The bridge window was closed, or the browser was quit and reopened
• Your browser blocked the pop-up when Claira tried to open the window
• The browser discarded or slept the bridge tab while it sat in the background
• Your computer went to sleep

How to resolve:

1. Click Open bridge (in Settings, or in the Local model badge) and leave the window open while you work.
2. If nothing opens, allow pop-ups for Claira, or use the link Claira offers instead.
3. A bulk scan that stopped this way is resumable. Open Bulk Tasks and click Resume Task; documents that already have results are not scanned again.

Your local model isn't responding.

The request left your browser, but your model server did not answer it.

Common causes:

• The model server is not running
• The server is running on a different port from the one saved for your local model
• Most commonly, the server does not allowlist the Claira address, so your browser refuses the request before it ever reaches the model; that setting is called CORS

How to resolve:

1. Confirm the server is up: open its own address in a browser tab and check that it answers.
2. Check your server's CORS setting. Each server spells it differently. Supported local servers gives the setting for Ollama, LM Studio, vLLM and llama.cpp. Allowlist the Claira address exactly as the Local model panel shows it, and not your model's address.
3. Restart the server after changing that setting: most read it only at startup.
4. Click Detect local servers again to confirm Claira can see it, then run the scan.

Your browser did not allow Claira to reach your local network.

Chrome asks for a local network permission the first time a website contacts a server on your own computer or network. Until you grant it, nothing reaches your model.

Common causes:

• The permission prompt was dismissed, or answered with Block
• The permission was reset or revoked later in Chrome's site settings
• A policy on a managed device blocks it

How to resolve:

1. Choose Allow when Chrome asks.
2. If you already answered, reset the permission in Chrome's site settings for the Claira address, then try again.
3. On a managed device, ask IT to allow the Claira origin with the LocalNetworkAccessAllowedForUrls policy. The policy names the Claira address (the address Claira is served from), and not your model's endpoint: what is being allowed is the page making the request, not the destination. See The browser permission.
4. Firefox and Safari have no equivalent prompt, so this error does not appear in those browsers.

That feature is not available with a local model.

Three things do not run on a local model, and this code covers all three: background scans; image, audio, video and PDF scans; and early case assessment. Nothing was sent to any model and nothing was charged. The background and non-text Scan as options are not offered, and early case assessment explains itself in its own words, so you normally see the explanation before you ever see the code.

Common causes:

• A bulk scan was started as a background task from a window that stayed open
• A saved scan configuration carried a media mode over from another case
• An early case assessment was run while your local model was on

How to resolve:

1. For a bulk scan, run it live instead, and keep the browser window and the bridge window open until it finishes. Background processing is not available in Claira Private.
2. For image, audio, video or PDF documents, set Scan as to Text.
3. Past early case assessments stay available to read. See Early case assessment.
4. What works with a local model lists everything that does run locally. It is a short list of exclusions, not a long one.

Your local model is not on for this case.

You should not normally see this one. It means your browser tried to record local work for a case where your local model is no longer on, usually because it was turned off somewhere else while this window stayed open.

Common causes:

• Your local model was turned off in another tab, or in another browser, while this window was still open
• The window sat open long enough for the case's settings to change underneath it

How to resolve:

1. Reload Claira. The panel picks up the current setting for the case.
2. If you want to keep working on your local model, turn it back on in Settings, under Your model, and confirm it in this browser.
3. A local model is yours and per case, so confirming it in one browser does not confirm it in another. Expect to turn it on again in each browser you work from.

Claira could not confirm whether your local model is on, so nothing was sent.

Claira checks your local model setting before any request that could leave your browser for a model. If that check itself fails (a momentary service hiccup, usually), Claira refuses the request rather than guessing. It never assumes your local model is off, so a hiccup can never send a document elsewhere while your panel shows your local model on. Nothing was sent and nothing was charged.

Common causes:

• A brief interruption on Claira's side
• A deployment in progress on Claira's side

How to resolve:

1. Try the action again in a moment. Bulk scans retry these documents on their own.
2. If it keeps happening for more than a few minutes, contact support.

Claira could not confirm who is asking, so it could not confirm your local model.

A local model is yours and per case, so Claira needs to know which user and which case a request belongs to before it can tell where the request may go. When it cannot tell, it refuses rather than proceeds. Nothing was sent and nothing was charged.

Common causes:

• Your Nuix Discover session expired while Claira stayed open
• Claira was opened outside its usual place in Nuix Discover

How to resolve:

1. Reload Claira inside Nuix Discover and try again.
2. If it persists, sign out of Nuix Discover, sign back in, and open the case again.

This request timed out on Claira's side after 30 minutes without progress.

For the work where Claira prepares the instructions and your model answers them (Insights, the Prompt Generator, Revise with Agent, Case Context from documents and Agent Mode), Claira holds the prepared request for 30 minutes while it waits for your model. After that it is dropped.

Common causes:

• The model is very slow on a very large job
• The bridge window closed part-way through
• Your computer went to sleep while the model was working

How to resolve:

1. Start the request again. Nothing about your case or your prompt has changed.
2. If it keeps happening, work on a smaller selection of documents, or load a faster model.

At least one selected document does not fit your model's context window even on its own.

For an Insight over a large set of documents, Claira reads them in batches and merges the findings, so a selection that is too big for a single pass is not a problem. A single document that will not fit at all is different, because there is nothing left to split it into, so Claira says so rather than quietly cutting the document short.

Common causes:

• One very long document in the selection
• A local model with a small context window; local models usually have much smaller windows than hosted ones

How to resolve:

1. Narrow the selection so the oversized document is not in it.
2. Open that document in Nuix Discover and use Unitize to split it into smaller parts.
3. Load a model with a larger context window, then click Re-run capability checks so Claira measures it again.

That needs a Claira Private case.

Connecting your own model, either as a shared connection or as a local model, is part of Claira Private. This case is on a standard plan, so the request was refused.

Common causes:

• A window that stayed open from before the case's plan changed
• A saved setting from a Claira Private case carried over to a case on a standard plan

How to resolve:

1. Reload Claira. On a standard plan, the case uses the models in the Model picker.
2. To run this case on your own model, deactivate it and activate it again on Claira Private. See Choosing Claira Private.

Claira's models are not used on a Claira Private case.

The request asked for one of Claira's own models, but this case runs only on your own model. Nothing was sent.

Common causes:

• A window that stayed open from before the case was activated on Claira Private
• A saved prompt or scan setting that names one of Claira's models

How to resolve:

1. Reload Claira and run the action again: it uses the shared connection, or your local model if you have one turned on.

Background processing is not available in Claira Private.

A bulk scan was started, sent or resumed as a background task on a Claira Private case. The task was not created or changed.

How to resolve:

1. Run the bulk scan in your browser, and keep the window open until it finishes.

No model is connected to this case.

A Claira Private case runs only on your own model, and this feature found none to run on: the case has no shared connection, and you do not have a local model turned on. The message names the feature that was refused.

Common causes:

• Nobody has set up the shared connection yet
• The shared connection was revoked
• You turned your local model off, and the case has no shared connection

How to resolve:

1. Open Settings and set up the shared connection in the Shared connection section under Team settings, or turn on a local model for yourself in the Your model section.
2. Run the feature again.

That feature is not available on a Claira Private case.

Claira cannot run this feature on your own model, and on a Claira Private case it never uses one of its own models instead. The message names the feature. Nothing was sent.

Common causes:

• Scan as was set to Image, Audio, Video or Auto: Claira Private scans text only
• A feature that Claira cannot run on your model, such as an early case assessment where it is not available

How to resolve:

1. For a scan, set Scan as to Text and run it again.
2. What Claira Private includes lists what runs on your model.

A case cannot switch between a standard plan and Claira Private.

The plan change was refused. A case moves between a standard plan and Claira Private only by being deactivated and activated again.

How to resolve:

1. Deactivate the case, then activate it again and choose the other option. See Choosing Claira Private.

Claira Private has no tokens.

A token action, such as claiming free tokens or buying a token pack, was attempted on a Claira Private case. Nothing was claimed or bought, and nothing was charged. A Claira Private case needs no tokens: it has one flat monthly price and no document limits.

How to resolve:

1. Nothing to do: carry on scanning. See Pricing and billing.

Claira could not confirm this case's plan, so nothing was sent.

Before any request that could reach a model, Claira checks whether the case is on Claira Private. If that check itself fails (a momentary service hiccup, usually), Claira refuses the request rather than guessing. Nothing was sent and nothing was charged.

Common causes:

• A brief interruption on Claira's side
• A deployment in progress on Claira's side

How to resolve:

1. Try the action again in a moment.
2. If it keeps happening for more than a few minutes, contact support.