Kanchi

Kanchi is a self-hosted Celery operations tool built for the part after an alert fires. It is aimed at operators who need to inspect failed work, choose a recovery path, keep an audit trail, and automate repeat incidents in Celery task workflows.

The product keeps failed, orphaned, retrying, and running tasks visible through live broker events, worker heartbeat context, and filters that match incident investigation. Task detail pages collect args, kwargs, traceback, queue, worker, retry chain, rerun lineage, and progress steps in one place. For recovery, bulk actions open a rerun review where payloads can be checked, inputs repaired, unsafe items skipped, and the outcome recorded. Kanchi also supports guarded automation with triggers, conditions, Slack or webhook notifications, retries, cooldowns, and circuit breakers, so repeat incidents can be handled without creating retry loops. Manual resolve, unresolve, rerun, skipped, failed, and edited actions are stored for later review, and long-running jobs can report progress percentages, named steps, and current messages.

Kanchi runs beside Celery workers and can point at RabbitMQ or Redis, with task operations added without changing application code. It stores task state, retry chains, action history, progress, and registry metadata in SQLite, Postgres, or MySQL. The interface also includes production access controls such as basic auth, OAuth, sessions, host and origin controls, and email allowlists. Prometheus metrics are supported. The project is self-hosted and MIT licensed.