Troubleshooting
Email issues
Section titled “Email issues”My email to the agent bounced
Section titled “My email to the agent bounced”Cause: Email address doesn’t exist or is misspelled.
Fix: Verify the agent name matches exactly. Check your dashboard for the correct email address.
- Free tier:
{name}-{hash}@tinyfat.com - Tiny tier:
{name}@tinyfat.com
Agent never responded
Section titled “Agent never responded”Possible causes:
- Your email isn’t in allowed senders
- API key is missing or invalid
- Job failed during processing
Fixes:
- Check allowed senders in Dashboard → Settings
- Verify API key in Dashboard → Settings → API Keys
- Check job history in Dashboard for error messages
Response went to spam
Section titled “Response went to spam”Fix: Add your agent’s email ([email protected]) to your contacts or mark as “not spam”.
Emails not threading properly
Section titled “Emails not threading properly”Cause: Email client or corporate server modifying headers.
Fix: Try a different email client. Gmail and Outlook usually work well.
Response took too long
Section titled “Response took too long”Cause: Complex requests take more time. The agent runs one turn per email.
Expected times:
- Simple questions: 10-20 seconds
- Tool use (web search, file ops): 20-60 seconds
- Complex reasoning: 30-90 seconds
If consistently slow, try /model haiku for faster responses.
API key issues
Section titled “API key issues””Invalid API key” error
Section titled “”Invalid API key” error”Cause: API key is malformed, expired, or has insufficient credits.
Fixes:
- Verify key at console.anthropic.com
- Check you have credits available
- Re-enter the key in Dashboard → Settings → API Keys
Usage not showing in Anthropic console
Section titled “Usage not showing in Anthropic console”Cause: It can take a few minutes for usage to appear.
Fix: Wait 5-10 minutes and refresh the Anthropic dashboard.
Session issues
Section titled “Session issues”Context too large / hitting token limits
Section titled “Context too large / hitting token limits”Symptoms: Agent mentions context limit, responses get cut off, high costs.
Fixes:
- Send
/compactto compress context - Send
/clearto start fresh - Switch to a longer-context model (all current models support 200k)
Lost conversation history
Section titled “Lost conversation history”Cause: Started a new session or used /clear.
Fix: Send /resume to see available sessions, then /resume {hash} to continue one.
Can’t find old session
Section titled “Can’t find old session”Cause: Sessions are stored for a limited time.
Fix: Important context should be saved to MEMORY.md for permanent storage.
Model issues
Section titled “Model issues”Wrong model being used
Section titled “Wrong model being used”Cause: Model preference saved incorrectly.
Fix: Send /model to check current model, then /model haiku|sonnet|opus to change.
Model not available
Section titled “Model not available”Cause: Your API key may not have access to all models.
Fix: Check your Anthropic account tier. Some models require higher tiers.
Dashboard issues
Section titled “Dashboard issues”Can’t log in
Section titled “Can’t log in”Fixes:
- Check spam for magic link email
- Make sure you’re using the same email you signed up with
- Request a new magic link
Settings not saving
Section titled “Settings not saving”Cause: Browser caching or network issues.
Fix: Hard refresh (Ctrl+Shift+R), try a different browser, check network connection.
Slack issues (Tiny tier)
Section titled “Slack issues (Tiny tier)”Bot not responding to @mentions
Section titled “Bot not responding to @mentions”Fixes:
- Verify bot is invited to the channel:
/invite @yourbot - Check event subscriptions are configured in Slack app
- Verify bot token is entered in Dashboard → Settings
Duplicate messages
Section titled “Duplicate messages”Cause: Slack retrying webhook delivery.
Fix: Usually resolves itself. If persistent, check Slack app webhook configuration.
Cost issues
Section titled “Cost issues”Unexpected high costs
Section titled “Unexpected high costs”Check:
- Job history for number of turns
- Token usage in session footers
- Model being used (Opus is 15x more expensive than Haiku)
Reduce costs:
- Use
/model haikufor simple tasks - Use
/compactregularly to reduce context size - Don’t send unnecessary messages
Cost not matching Anthropic dashboard
Section titled “Cost not matching Anthropic dashboard”Note: TinyFat shows estimated costs based on token counts. Actual billing comes from Anthropic and may differ slightly due to:
- Caching discounts
- Rounding
- Timing of usage reporting
Getting help
Section titled “Getting help”If you’re still stuck:
- Check job history — Dashboard shows errors for each job
- Email support — [email protected]
- Include details:
- Your agent’s email address
- Timestamp of the issue
- Any error messages you see