Trigger.dev runs background work that outlives a request: durable chat agents with tool calling and human-in-the-loop, scheduled cron tasks with no timeout ceiling, concurrency and queue control, and realtime streams that push LLM output straight to a frontend. Python, Prisma, Puppeteer, FFmpeg and apt-get are all available in the runtime.
No timeout is the specification that matters. Serverless functions cap out at minutes, agent runs and video processing do not, and the usual workaround is chunking work into pieces that fit the platform rather than pieces that make sense. Realtime streams solve the adjacent problem of showing a user what a long task is doing while it runs.
It is Apache 2.0 with more than 16,000 GitHub stars, built by API Hero, with observability, tracing and filtering included. Pricing is described as pay for what you use with no figures published, usage-based billing on long-running compute needs watching, and self-hosting the open-source version is real operational work.
