Trigger.dev is an open-source TypeScript platform for running long-running background tasks, AI workflows, and agents. You write tasks as exported functions in your own codebase with the SDK, and the platform runs them with no execution timeouts, durable checkpointing, automatic retries, queues, and elastic scaling. It runs as a managed cloud service or self-hosted with Docker Compose or a Kubernetes Helm chart.
For AI agents, it works alongside the frameworks, services, and models you already use. Tasks can pause for human approval or feedback through waitpoints, stream LLM responses to a frontend in real time, and be subscribed to from React hooks. Durable chat agents survive page refreshes, redeploys, and crashes, and tool calls can be marked as needing approval while the run stays alive.
Deployed tasks can be customized with system packages through build extensions, so a task can run browsers, Python scripts, or FFmpeg. Each run has full tracing and logs, and alerts for failed runs or deployments go out by email, Slack, or webhook. Separate development, preview, staging, and production environments, atomic versioning, and per-task machine sizing support moving tasks from local testing to production.
Features
- Long-running tasks: no execution timeouts, with checkpoint and resume for durability
- Retries and queues: automatic retries, idempotency, concurrency rules, and queues
- Human-in-the-loop: waitpoints pause a task until a person approves, rejects, or responds
- Realtime and streaming: subscribe to runs and stream LLM output to your app, including React hooks
- Durable cron schedules: recurring schedules attached to tasks
- Build extensions: add system packages to run browsers, Python scripts, FFmpeg, and more
- Structured inputs and outputs: schema-defined task payloads with runtime validation
- Batch triggering and waits: start many runs with custom payloads and pause execution for set durations
- Environments and versioning: development, preview, staging, and production with atomic version deploys and preview branches
- Observability: tracing, logs, tags, live run metadata, bulk replay or cancel, and failure alerts
- Machines: configure vCPUs and memory per task
- Deployment options: managed cloud or self-hosted with Docker or Kubernetes
