Error codes and fixes
Wording may vary — match symptom → where to check. UI anchors: Settings / Extensions / History / Chat.
API · auth
API key is invalid / API key is missing
Check: Settings → AI & Models → Provider API Keys → re-enter key → Save Changes
Re-issue keys in the provider console (Anthropic / OpenAI / Google AI Studio, …).
Rate limit exceeded
Wait and retry. Switch Default LLM or the session Model, or reduce concurrent sessions / scheduled tasks.
401 / Authentication failed
Re-paste the key without spaces. For MCP, re-check env/auth fields under Extensions.
Connection · network
Connection refused
Confirm Ollama / local MCP is running and Base URL / port are correct.
MCP: verify command · args · cwd in Extensions, or run the same command in a terminal. → Custom MCP
Timeout / Request timed out
Narrow the task, check network / MCP latency, then retry. Long jobs: split with Sub-agents.
MCP not responding
Extensions → server config → dependencies (npx / uv, …) → remove and re-add. → Extensions
Sessions · models · tools
Session not found
Confirm in History. If deleted, start a new session from Chat.
Model not found / Model unavailable
Default LLM in Settings, or Chat Model picker → Refresh models. Also check the assistant’s default model.
Tool execution failed
Paste the tool error back to the agent and ask it to diagnose. Check workspace paths, OS permissions, and approval / YOLO state.
Context window exceeded
Start a fresh session and paste a short summary if you need continuity. Avoid attaching huge files or long logs in one session. You can also lower Max Input Context under Chat Interface.
Files
Permission denied / File not found
Check the session work-folder path and OS permissions. Child sessions use a separate folder by default. Ask them to reuse the current work folder when they must see the same deliverables — Sub-agents.
Related
Support: Discussions — no API keys.