Postgres-backed orchestration for tasks, DAGs, and durable workflows

What it does

Hatchet orchestrates background tasks, AI agents, and long-running workflows for applications written in Python, TypeScript, Go, and Ruby. It bundles queuing, automatic retries, cron scheduling, event triggers, and a real-time web UI into one platform that can be self-hosted or run as a managed cloud service. Workflows range from simple fire-and-forget functions to complex directed acyclic graphs and fault-tolerant durable tasks that recover from crashes.

The interesting bit

Rather than adding Redis or RabbitMQ to your stack, Hatchet uses Postgres as its durability layer for both task runtime and observability. That keeps the infrastructure footprint small for self-hosting and turns the same database into the permanent source of truth for execution history, multi-tenancy, and metrics.

Key highlights

• Durable tasks with pause/resume via sleep and event waits, pitched as a drop-in replacement for Temporal or DBOS workflows.
• DAG support for data pipelines alongside general-purpose background queues, webhooks, and event-driven listeners.
• Fine-grained fairness controls: worker affinity, dynamic rate limits, concurrency limits keyed on dynamic values, and per-worker slot caps.
• Built-in observability via OpenTelemetry and Prometheus, plus multi-tenancy, user roles, and alerting in the web UI.
• Load-tested to 10,000 tasks per second, though the README notes it consumes more resources than lighter Redis-backed queues.

Caveats

• Durability has a cost: the README concedes higher resource usage than Redis or RabbitMQ brokers and cites a 10k tasks/second load-test ceiling.
• Data pipeline integrations are DIY; unlike Airflow or Prefect, Hatchet does not ship with a broad catalog of out-of-the-box datastore connectors.
• Ruby is listed as a supported language, but the README’s SDK examples and topics focus almost entirely on Python, TypeScript, and Go.

Verdict

Teams that want durable execution, fair scheduling, and multi-tenant background jobs in a single self-hostable platform should take a look. If you need plug-and-play data connectors or bare-metal queue throughput above 10k tasks/second, look elsewhere.