Troubleshooting

App Won't Start

Symptom: CoreLayer fails to launch or crashes immediately.

Solutions:

Verify Node.js 22+ is installed: node --version
Verify pnpm 9+ is installed: pnpm --version
Check platform-specific Tauri prerequisites
Try rebuilding: pnpm clean && pnpm build

Daemon Not Starting

Symptom: "Daemon unhealthy" or connection errors.

Solutions:

Check if port 3001 is in use: lsof -i :3001 (macOS/Linux) or netstat -ano | findstr 3001 (Windows)
Kill conflicting processes: pnpm clean:daemon
Check daemon logs in the Control Center
Verify database file permissions

Model Not Responding

Symptom: Jarvis doesn't respond or shows "No model available".

Solutions:

Verify API key is correct in Settings → Models
Check provider status (Groq, OpenRouter, etc.)
Try a different model/provider
Check network connectivity

MCP Server Won't Connect

Symptom: Server shows "Disconnected" or "Error" status.

Solutions:

Verify the command and args are correct
Test the command manually in a terminal
Check that npx can install the package: npx -y @modelcontextprotocol/server-filesystem
Check environment variables are set
Restart the MCP connection from the Control Center

Voice Not Working

Symptom: Wake word not detected or speech not recognized.

Solutions:

Check microphone permissions in system settings
Reduce background noise
Try clicking the microphone button instead of wake word
Switch ASR provider to Groq Whisper for better accuracy
Check audio output settings for TTS

Permission Guard Blocking Everything

Symptom: All tool calls require approval.

Solutions:

high risk always requires confirmation and critical is denied — this is the default policy, not a bug
Use "Remember" on the approval to stop being asked for that exact action
Clear stale requests with POST /api/approvals/expire-stale
The policy is not user-configurable yet; see Permissions

Approval Appears Stuck

Symptom: A run never finishes and no approval dialog is visible.

Solutions:

Check GET /api/approvals for a pending request belonging to the run
Approvals are persisted, so a daemon restart does not lose them; it also does not auto-decide them
Inline tool confirmations live at GET /api/tools/pending-confirmations

Coding Executor Fails Immediately

Symptom: A coding run fails before producing any output.

Solutions:

GET /api/runtimes/coding/diagnostics — check the agent probe state
needs_auth means the agent started but has no credentials; complete its login
unreachable usually means acpx is not installed (npm install acpx) or the launch command is wrong
Claude Code runs on DeepSeek models only; a non-DeepSeek ANTHROPIC_MODEL is rejected on purpose

Eval Attempts Time Out

Symptom: pass^k collapses with many timed-out attempts.

Solutions:

Check the gateway first: a flapping endpoint makes acpx hang until the run budget expires instead of failing fast
VOID attempts are environment faults (quota, refused connection, 5xx) and are excluded from rates
Three environment faults in a row stop the eval run by design

`POST /api/runs/:id/message` returns 410

This endpoint was removed with the hand-written agent loop. Use the standard conversation send API (POST /api/conversations/:id/messages/stream) instead.

Build Failures

Symptom: pnpm build fails.

Solutions:

Clean and rebuild: pnpm clean && pnpm install && pnpm build
Check Rust toolchain: rustc --version, cargo --version
Verify platform prerequisites for Tauri
Check for TypeScript errors: pnpm typecheck

Database Issues

Symptom: Data not saving or loading errors.

Solutions:

Check storage mode in storage.json (Settings → Storage writes it)
Verify the database file exists and is writable under storage/
Switching storage mode requires a daemon restart — it is not a live hot-swap
For Supabase: verify project URL and service key

Still Stuck?

Check the GitHub Issues for known problems
Open a new issue with:
- Your platform and versions
- Steps to reproduce
- Error messages or logs
- Expected vs actual behavior

On this page