Skip to content

Troubleshooting ​

Find your symptom in the index, check the likely cause, then follow the link to the fix. Start with the visible task state and error message: a slow upstream service, a queued task, and a disconnected browser need different responses.

Symptom index ​

SymptomLikely causeGo to
A search takes a long timeQueued task, large budget, or a slow upstream stageSearch problems
HTTP 401 / 403Wrong key or missing provider permissionHTTP errors
HTTP 404Wrong base URL, protocol, route, or modelHTTP errors
HTTP 429Provider rate limit or insufficient creditHTTP errors
Connection error / timeoutNetwork reachability or provider availabilityHTTP errors
Your endpoint is rejected onlineNon-public address (localhost, private network)My endpoint is rejected online
No results, missing abstracts, or counts differOver-restrictive query or gaps in the source recordNo results
The citation map is emptySource without expansion, or no links within budgetThe citation map is empty
A review outline or section failsModel availability or a provider errorA review outline or section fails
The local service cannot startPython version, database path, or port conflictLocal service cannot start
Sign-in or email-code problemsSpam folder, expired or replaced codeYour account

Search problems ​

A search is taking a long time ​

Check whether the task is queued or running. During execution, query refinement, source requests, citation expansion, and model scoring can each take time. Larger budgets make more work possible but do not guarantee more useful results.

If the page loses its connection, reopen the saved run and inspect its status before submitting again. Anonymous online results are temporary; see Data & privacy.

No results, missing abstracts, or different counts ​

Broaden the question, remove overly restrictive fields, or try a more suitable source. Source hits are not the number downloaded, and downloaded candidates are not the final ranked set.

Missing abstracts, citation counts, or PDFs can be a property of the source record. PaperSeek does not independently fill missing abstracts from another source.

The citation map is empty ​

Check that the run used a supported citation-expansion source and that expansion was enabled. Even then, the source may provide no usable links within the available budget.

HTTP and provider errors ​

HTTP 401, 403, 404, or 429 ​

ErrorCheck
401 / 403Correct key, provider account permissions, and selected model
404API base URL, protocol, route, model availability, or a missing task/resource
429Provider rate limit or structured insufficient-credit error
Connection / timeoutNetwork reachability and provider availability

A 404 displayed during an OpenAlex run does not by itself prove that OpenAlex returned it: model requests and application routes also occur during the task. Record the stage, safe error code, task ID, and time.

PaperSeek Service credits and an upstream provider's balance are different. Increasing one does not change the other.

My endpoint is rejected online ​

Online, your own key cannot point to localhost, private networks, and other non-public destinations. Use a public provider URL, or use Community for a model running on your own machine. A valid URL must also implement the chosen API protocol.

Review problems ​

A review outline or section fails ​

Review writing is experimental and depends on selected evidence and model availability. Select papers with useful abstracts, inspect the current project/job state, and retry the failed work after resolving the reported provider issue. Avoid repeatedly creating duplicate projects.

Longer drafts need more output from the model. A completed outline does not guarantee that the next writing request will succeed. Keep manual edits and inspect versions before regenerating.

Local service ​

Local service cannot start ​

Run paperseek doctor --json. Check Python 3.12+, the selected database path, write permissions, port conflicts, and model settings. Read the launching terminal or service manager's logs; the Web UI intentionally focuses on status rather than raw execution logs.

For separate processes, confirm both API and worker use the same database.

Report a reproducible issue ​

Open a GitHub issue with:

  • Installed version or online mode, browser/OS, and steps to reproduce.
  • Source, protocol, ranking method, and budget.
  • Expected behavior, observed error, task ID if appropriate, and approximate time.
  • A minimal non-sensitive example and a cropped screenshot if helpful.

Remove API keys, passwords, verification codes, cookies, private queries, and identifying data. Security reports belong in the repository's security process.

Open source. Open to exploration. Apache-2.0 licensed.