Docs / Reference
REFERENCE / AI POD

Troubleshooting

Common causes of blocked work, missing evidence and setup issues.

Updated October 9, 2026 · 5 sections

The Pod is blocked

Check provider status, company blueprint version, task dependencies and pending decisions. A live Pod needs a compatible chat gateway. Fix the cause, then retry the blocked Pod. An expired worker lease also blocks for explicit recovery.

A file has no searchable text

Images and XLSX files are stored but not extracted. A scanned PDF needs OCR. Check the file’s extraction status and upload a readable source where possible.

A tool does not run

An onboarding tool selection or discovered MCP schema does not enable execution. Check integrations and whether a dedicated adapter exists.

No verification email arrives

Both local and production require RESEND_API_KEY and a verified sender in RESEND_FROM. Check spam, the sender domain verification and delivery status in Resend. Restricted test senders cannot deliver to arbitrary work emails. Wait 60 seconds before resending. Codes are not printed in logs. If delivery fails, request a new code.

Work does not start in production

Verify the persistent worker process is running against the same database as the web process. npm start serves HTTP; it does not start the worker.