Reference

Fix model, upload, and knowledge search problems

Check an unavailable model, rejected document, failed processing job, or weak knowledge result.

Start with what you see

Choose the closest problem

Select one symptom. You will still check the evidence before changing anything.

What you see

  • No approved model is available for the required work.
  • A model file upload stops or the local runtime is unavailable.
  • A document is rejected or stays failed in Operations.
  • Knowledge search returns no passage or the wrong passage.
A knowledge database with model, document, and search areas
Check the embedding model, document state, and test search on the same page. Earlier development interface, captured 22 August 2026. Follow the written instructions for 1.0.0. Open the full image.

What it means

The required model may not be ready or enabled. A document must pass the file policy, be readable, and finish processing before its passages can be searched.

Search also needs the exact Search action. View access is not enough.

Likely causes

  • The model Connection, deployment, local runtime, or model file is not ready.
  • The document type or size is not allowed.
  • The file is empty, encrypted, damaged, or contains no readable text.
  • Document processing is paused, waiting, or failed.
  • The question does not use words or meaning found in the stored passage.
  • The person lacks Search access to this knowledge database.

When an external model Connection test times out

For the reviewed 1.0.0 source follow-up, an expired provider probe deadline is marked provider_timeout in the correlated operational diagnostic record. Use the request ID and time to find that record, then check the provider address, reachability and configured timeout before retesting once. This category is diagnostic log context, not a new public API error code. A cancelled request or a provider rejection is a different cause. This check does not apply to the bundled local model runtime.

Safe checks

  1. Check the model state and the work it is enabled to perform.
  2. Test the model Connection without copying its secret.
  3. Check the current upload policy before trying the file again.
  4. Open the document and read its processing stage and message.
  5. Find the matching job in Tenant operations.
  6. Use Check access for Search on the named knowledge database.
  7. Search for one fact that clearly appears in a finished document.

Solutions

  • Enable an approved ready model for the required work.
  • For an interrupted GGUF upload, try again or place the approved file in the configured model folder and refresh.
  • Use a supported file that is within the shown limits and contains readable text.
  • Resume document processing when an authorised administrator paused it.
  • Correct the identified processing cause before retrying the job.
  • Add Search access only when the person should search the database.
  • Improve the source or search question when the stored passage does not contain the needed information.

What to collect

Collect the tenant, model or database name, safe file name and type, file size, processing stage, time, job ID, request ID, and visible message.

Do not collect model credentials, provider keys, customer document contents, or a private model file.

Read Prepare a model and knowledge database, Upload and search knowledge, and Run local AI models.

The model is loaded but GPU work fails

Check the supplied runtime, selected processor, memory, original failure and any recovery wait. Use the worker recovery guide. Loading again alone does not prove a successful workload.

Processing is waiting or stopped

Read the document stage, current owner, capacity wait, and original error. Use document controls and model scheduling.

Runtime in the 1.0.0 packages

The published Linux packages supply CPU inference and embedding. They contain no GPU worker runtime. GPU policy controls describe what the source can do with a suitable runtime; they do not add one to the installed package. The published Windows MSI includes a CPU inference runtime. Confirm any GPU runtime against the exact installed package. Check the package runtime scope before choosing a backend.

Pūnaha Docs

Search the guides

Enter at least two characters.

    Product screen

    View the full screenshot