Troubleshooting

Start with System → Status (admin): it checks the media server, the AI provider, the lemmatiser and each language's model, and links to the screen that fixes each problem. docker logs learnarr shows the rest.

Sign-in fails, or the page keeps returning to sign-in

• Behind a reverse proxy with HTTPS, set PUBLIC_URL to the address you use.
• Reached at an address Learnarr doesn't expect (a second hostname, a VPN address), add it to TRUSTED_ORIGINS.
• Over plain HTTP, don't set PUBLIC_URL to an https:// address: browsers drop secure cookies on HTTP.

"Your media server has no subtitle for this"

The episode has no text subtitle track in your language. Image-based tracks (PGS, VobSub, common on Blu-ray rips) can't be read. Connect your OpenSubtitles.com account in Settings → Account to download one, upload a file, or have Bazarr fetch one.

OpenSubtitles says the download limit is reached

Free OpenSubtitles.com accounts have a daily download limit. It resets every 24 hours. Episodes you already have never count again.

No "% you know" score

The episode page says why under the score:

• Downloading the … language model: wait; it's once per language.
• Couldn't reach lindat.mff.cuni.cz: the server can't download models. Allow outbound HTTPS to that host, then press Try again.
• There's no language model for …: Learnarr can't lemmatise that language yet (see Languages).
• The lemmatiser isn't installed: only outside Docker; run scripts/fetch-udpipe.sh.

AI suggestions fail

• returned an error (HTTP 401): the API key is wrong or has no credit.
• ran out of room before answering: a reasoning model spent its answer on thinking. Pick a non-reasoning model.
• didn't answer with JSON or didn't match any subtitle line: the model is too small or ignores instructions. Use a larger one.
• Timeouts with a local model: the first request loads the model, which can take a minute. Try again.

Notifications don't arrive

See Notifications.

Reporting a bug

Open an issue with what you did, what you expected and what happened, the Learnarr version (System → Status) and the relevant lines of docker logs learnarr. Remove anything private (tokens, addresses) first.