Error Cockpit

If you see a red or yellow ring: That service is experiencing failures right now.

• Most likely: The external API (Gemini, Deepgram, Google) is having an outage or rate-limiting us.
• Also possible: A recent code change broke how we call that API (wrong params, missing auth).

What to do:

• Click the card to filter the Recent Errors table below — read the error messages to diagnose.
• Check the provider's status page (e.g. status.cloud.google.com) to see if it's on their end.

Free tier limits to watch:

• Supabase: ~30 signups/min, 60 DB connections, 4 auth emails/hr, 500K edge invocations/mo, 500MB DB. If supabase is red, it may be time to upgrade to Pro ($25/mo).
• Deepgram: 600 min/month per user, ~10 concurrent WebSocket connections, ~3GB max file upload. If deepgram is red, check if users are recording very long meetings (>4 hours). voice_transcribe errors indicate voice input issues in coach chat (auth, network, or empty recordings).
• Audio Capture: If audio_capture is red, check for native CLI errors (permission denied, missing binary) or disk space issues. Common error codes: SPAWN_FAILED, EXIT_CODE_1, LOW_DISK_SPACE, EMPTY_AUDIO_FILE.
• Gemini: Rate limits vary by model. If gemini is red during peak hours, multiple users may be analyzing meetings simultaneously. generate_prep_flashcards errors = meeting prep feature (uses gemini-2.0-flash, NOT 2.5-flash which produces empty output).
• Resend: Email delivery for meeting prep flashcards. Free tier: 100 emails/day. send_prep_email errors = email delivery failures (check Resend dashboard for bounces/complaints). If resend is red, check RESEND_API_KEY Supabase Secret.
• Google Calendar: 1M API calls/day per GCP project.
• Outlook Calendar: Microsoft Graph API. oauth_connect errors = user's org may block third-party app consent (common AADSTS65004). token_refresh errors = session expired or revoked. OAUTH_TIMEOUT = user didn't complete auth within 5 min. Web app uses outlook-oauth-exchange Edge Function for token exchange. If Outlook errors spike, check if OUTLOOK_CLIENT_SECRET_WEB is set in Supabase Secrets and if the Azure app registration is still active.
• Exa (Learning Feed): $10 free credit (~2500 searches at $0.004/search). generate_feed errors = feed generation failed (Exa search + Gemini curation pipeline). Common error codes: NETWORK_ERROR (can't reach Exa/Gemini), RATE_LIMITED (API quota), TIMEOUT (Edge Function took too long, typical >30s), SQLITE_SAVE_FAILED (local DB write error). Auto-refresh errors (AUTO_REFRESH_ANALYSIS_FAILED, AUTO_REFRESH_DIGEST_FAILED) are non-critical — user can still manual refresh. If exa credit exhausted, check dashboard.exa.ai billing.

If you see a spike: Many API calls failed during that time window.

• Most likely: An external API had a temporary outage or we hit a rate limit (e.g. Supabase free tier auth limits, Google Calendar API quota).
• Also possible: Multiple users signed up or connected calendars simultaneously, overwhelming shared resources.

What to do:

• Hover over the spike to see which services were affected. A single-service spike = that provider had issues. Multi-service spike = likely our network or Supabase was down.
• If yellow (rate limited) bars dominate, you're hitting free tier limits. Supabase free tier: ~30 signups/min, 60 DB connections. Consider upgrading to Supabase Pro ($25/mo) or adding request throttling.

If you see a latency spike: API calls to that service were abnormally slow during that window.

• Most likely: The provider was under load (Gemini thinking models can take 10-30s during peak). Large payloads (long meeting transcripts) also increase latency.
• Also possible: Network issues between the user's machine and the API, or the user was on a slow/unstable connection.

What to do:

• If Gemini latency is consistently >15s, check if we're sending oversized prompts or if the model tier has changed.
• If Deepgram latency spikes, check if audio files were unusually large (long meetings without live streaming).

If you see repeated errors here: The same API call is failing multiple times.

• Most likely: A persistent issue — wrong API key, expired token, or the provider changed their API (breaking change).
• Also possible: A transient issue that's already resolved — check timestamps. If the last error was hours ago and health is green, it's likely fixed.

What to do:

• Look at the error_code and message columns. Common patterns: "401" = auth issue (check API keys), "429" / "RATE_LIMITED" = hitting free tier limits (consider upgrading), "CONNECTION_POOL_FULL" = too many simultaneous DB queries (Supabase free tier: 60 connections), "PGRST116" = missing database row (profile backfill needed), "email not confirmed" = user signed up but never confirmed email (check Supabase email delivery or disable confirmation).
• Filter by service to isolate the problem. If it's supabase/profile_create — a user may have signed up but their profile wasn't saved. If you see many sign_up successes followed by sign_in errors with "email not confirmed", the Supabase email provider may be rate-limiting (free tier: 4 emails/hr).

If you see an investigation here: The system detected a repeated error pattern and automatically analyzed it using Gemini.

• Most likely: A real issue that needs attention — the same error happened 3+ times in 10 minutes, which rules out one-off glitches.
• Also possible: A transient issue (marked with "transient" badge) that resolved itself — e.g. a brief API outage.

What to do:

• Read the root cause and recommended action. If severity is "critical", act immediately. If "warning", investigate within 24 hours.
• If no investigations appear but errors are present above, the service may not be monitored yet — check MONITORED_SERVICES in limitsMonitor.js.