Edit on GitHub

AI problems

The AI section of Administration → Instance shows the Queue: how many files are waiting, running and failed, and how much space the models use. The server output shows errors from AI jobs.

Models don't download

The models come from Hugging Face the first time they are needed and are kept in data/models (up to 3 GB). Without internet access, or with huggingface.co blocked, jobs fail and retry.

  • Allow outgoing HTTPS to huggingface.co and its download hosts.
  • Or copy data/models from another installation with the same preset.
  • Check free disk space.

Out of memory

The server, or the Docker container, is killed (exit code 137, or Killed in the log) while files are analyzed. The models run inside the Fileship process, so the whole process restarts.

  • Choose a smaller Preset: Low (4 GB RAM), Medium (6 GB RAM) or High (8 GB RAM) are the memory each needs.
  • Keep Parallel jobs at 1 on small servers. Each parallel job needs its own memory.
  • Turn AI features off if the server has 2 GB or less.

After a crash, the running jobs are picked up again after Job heartbeat timeout; a file that fails three times is marked failed.

The server is slow while files are analyzed

Analysis runs on the CPU. Lower Threads per job and Parallel jobs so the dashboard keeps some cores.

The queue is stuck

  • Waiting files but nothing running: AI is off for the instance, or the files' owners turned it off in Settings → AI.
  • Running for a long time: a large video or PDF takes a while. A job whose process died is picked up again after Job heartbeat timeout (2 minutes by default).
  • Failed files wait for Retry failed. Failed jobs are cleared after 30 days, and those files are then tried again.

Search doesn't find a file

  • While a file is still being analyzed, only full-text search by name finds it.
  • Smart search only shows results that clearly match, so vague queries can return few or none. Try a more specific description, or turn off Smart search to search names.
  • After changing the preset, files are analyzed again in the background; results improve as it finishes.

AI features are turned off

You used an AI feature (such as text capture in the Windows app, which uses OCR) while AI is off for the instance or for your account. A super admin turns it on in Administration → Instance → AI, and you in Settings → AI.

A file was wrongly marked as sensitive

Choose Mark as not sensitive from the file's menu. Your choice is kept when the file is analyzed again. See Sensitive content.