Trigger.dev
Open source platform for durable AI agents and long-running tasks
About Trigger.dev
Trigger.dev is an open source platform for writing long-running background tasks in TypeScript and deploying them without managing infrastructure. The idea it is built around is durable execution: a run is checkpointed as it goes, so a crash, a redeploy or a browser refresh does not lose the work, and the task picks up where it stopped rather than starting again. Runs have no maximum duration at all. The company's own FAQ puts it plainly: you set a default maxDuration and can override it per task or per run, or set it to timeout.None to let a run execute indefinitely, and they say they have seen user runs last more than six months. That shape matters most for AI agents, which spend their time waiting on a model, then on a human, then on an API, and which a serverless function with a fixed timeout cannot hold.
The second idea is that waiting should be free. When a task triggers and waits on a subtask, or calls wait.for or wait.until for longer than five seconds, the process is snapshotted and restored later, so idle time counts against neither the compute bill nor the concurrency limit. The licence is Apache-2.0 and there is a published self-hosting guide, so the subscription buys managed infrastructure rather than the right to run the software. It is worth understanding the billing shape before comparing it to anything: each plan includes a monthly credit allowance, and the same metered compute consumes those credits, so the figure on the pricing page is a floor rather than a bill. The company behind the project is API Hero Ltd.
You write tasks as ordinary TypeScript functions inside a trigger folder in your existing project, commonly src/trigger, and the CLI bundles and deploys that folder. Locally you develop against the same runtime with hot reloading, and runs in the DEV environment are not charged. Tasks are triggered from your application code, on a schedule, or by a webhook, and each run gets a dashboard entry with logs, traces and a replay. Because tasks live in your own repository they are version controlled alongside everything else, and the dashboard can roll back to a previous deploy.
Usage is billed in two ways on managed workers. Compute is charged per second at a rate set by the machine size, from $0.0000169 per second on Micro (0.25 vCPU, 0.25GB) to $0.00068 per second on Large 2x (8 vCPU, 16GB), and every run that starts executing is charged $0.000025 as an invocation cost, which the page restates as $0.25 per 10,000 runs. Concurrency limits apply only while code is actually executing. All plans are limited to 1,500 API requests a minute, and batchTrigger() packs up to 1,000 tasks into a single API call.
- •Durable Execution - Runs are checkpointed and restored, so they survive crashes, redeploys and refreshes instead of being lost and retried from the top.
- •No Maximum Duration - There is no platform timeout. maxDuration is yours to set per project, per task or per run, or to switch off entirely.
- •Checkpointed Waits - Waiting on a subtask, or on wait.for and wait.until beyond five seconds, snapshots the process, so idle time costs neither compute nor a concurrency slot.
- •Agent Primitives - chat.agent wires a system prompt, tool calling, tool approval for human-in-the-loop, compaction, mid-turn steering and telemetry. It works with the Vercel AI SDK and any LLM provider, and Realtime streams responses and tool calls to a frontend with no API routes of your own.
- •Observability and Replay - Logs, traces, custom dashboards and alert destinations, with retention set by plan at 1 day on Free, 7 on Hobby and 30 on Pro. Runs can be replayed individually or in batches.
- •Self-Hosting Under Apache-2.0 - The platform is open source with a self-hosting guide, so the managed service is a convenience rather than the only way to run it.
- •Python From TypeScript - A Python build extension installs Python and its dependencies at build time, so a TypeScript task can call into a Python script at runtime.
TypeScript and JavaScript teams building AI agents, long-running workflows, media processing, scheduled jobs or browser automation, and anyone who has watched a serverless function time out halfway through an agent loop. The free plan is genuinely usable for a side project at $0 with $5 of monthly compute credits and 20 concurrent runs, and the self-hosting route suits teams whose security review will not allow task code on someone else's infrastructure. Cal.com describes running millions of workflows a month on it.
It is a poor fit in three cases. Python-first teams get Python only as something a TypeScript task calls into, not as a first-class way to author tasks. Anyone who needs a predictable flat monthly bill will find the plan fee plus metered compute plus a per-invocation charge hard to forecast, and the soft limits on the Pro plan (200 concurrent runs, then $10/month per 50; 25 team members, then $20/month per seat) make a like-for-like comparison with a flat-rate competitor misleading. And teams that want a supported self-hosted deployment should read the FAQ first: it points at a self-hosting guide, not at a supported product.
Pricing
$0 - $50/mo
- Free$0/mo
- Hobby$10/mo
- Pro$50/mo
- EnterpriseContact sales
From the vendor pricing page, 2026-11-01
