Troubleshooting

Start with lookout doctor: one line per check, with the reason and a hint, and a non-zero exit if anything fails.

"Is Ollama running?"

The model endpoint refused the connection. Start it with ollama serve. If doctor says the endpoint answers but has no model, run the ollama pull it suggests.

Items stay "waiting for the model"

Either the daily limit is spent or the model is down. The run summary at the top of the dashboard says how many items wait ("12 waiting for the model"), and so does lookout run.

• The limit is spent. llm.daily_limit counts model calls per day (200 by default). The rest are read the next day, most urgent first, and rule-based priorities (a critical advisory on your stack, for example) already apply. A local model costs nothing, so raise the limit: daily_limit: 2000.
• Only items you can see are read. Done, dismissed and expired items are never sent to the model, so a big backlog does not eat the budget. Reopen one and it joins the queue.
• The model is down. Run lookout doctor.

"The model could not read it"

The model answered twice without valid JSON. Small models struggle with the format; try a larger one. The item stays in the inbox with a neutral priority.

A source shows an error

The Sources page and lookout doctor show the last error. Lookout retries with growing delays; fix the URL or wait. A page that is not a feed is reported as such.

"Cannot reach Lookout" in the dashboard

lookout serve is not running, or you opened a different port.

The dashboard is blank or 404 on /

The dashboard is not built. Run npm run --prefix frontend build, or point dashboard_dir at the built files.

A priority looks wrong

Open the item: Why this rank shows the rule or score behind it. If a rule should have applied, check the item names a package in your stack (whole word, or an alias), then run lookout reassess. To report a wrong model reading, open an "Eval case" issue.