Help
Signing in and finding sessions¶
| What you see | What to do |
|---|---|
| A disabled sign-in button | Use another enabled provider or contact the service operator. |
| An empty Sessions list | Check that you used the same account that created the session. |
| A missing model or distribution plan | Ask the service operator whether it is available. Only enabled choices appear. |
Getting a session ready¶
| What you see | What to do |
|---|---|
| Workers are waiting | Share the contributor invitation and make sure enough people join for the selected plan. |
| Workers are preparing | Check progress on their contributor pages. Loading model pieces can take time. |
| A contributor fails its GPU or memory check | Read the contributor’s message. Try a suitable device, or create a session with a plan whose per-contributor memory requirement it can meet. |
| Direct links are connecting | Allow up to 30 seconds after loading. Check Connection details for direct, TURN or fallback status. |
| Open model chat is unavailable | Wait for the coordinator and all required contributors to be ready. Check the session’s status message. |
| A contributor disconnects during use | Restore its connection and check readiness before retrying. If the session is failed or closed, create a new session. |
Keep contributor tabs open and computers awake. Your chat tab alone does not supply compute unless you also join through I’ll contribute myself.
API errors¶
Errors normally include an error.message explaining what went wrong.
| HTTP status / error | What to check |
|---|---|
400 — invalid request | Session ID format, exact model ID, supported fields and text message roles. The last message must be from the user. |
400 — output cap or context capacity | Lower the requested output cap or shorten the message history. Do not supply both cap fields. |
404 — session_not_found | Copy the Session ID again and use the coordinator endpoint shown on that session page. |
409 — session_not_ready | Return to the session page and wait for serving readiness. |
409 — session_busy | Another API request is using the session. Wait for it to finish before retrying. |
413 — request too large | Reduce the JSON body to at most 1 MiB. |
503 — inference_error | Inspect contributor readiness and the session message. Retry only after availability returns. |
Streaming failures can arrive as an error event after the HTTP response has
started. Check the stream body as well as the HTTP status. Do not automatically
replay a request while the previous reply may still be running.
Ask for help¶
Send your service operator the model name, time of the problem, displayed error and whether you were using chat or the API. For connection issues, include the connection mode shown on the session page. Share diagnostics privately: they may contain network addresses. Keep session IDs, invitation links and chat URLs out of public issue reports because they grant access.