the problem
Long-running work should not disappear because a process restarted, a network request timed out, or a task needed to sleep for a while. Most teams either build that machinery themselves or adopt a hosted orchestration system before they know what they need.
what it explores
durlo is an experimental TypeScript and PostgreSQL library for durable tasks and workflows. It persists step state, retries failed work, survives crashes, and makes long waits explicit instead of keeping a process alive.
The current work includes cancellation, manual retry, workflow checkpoints, durable sleeps, a local CLI, and an inspection dashboard. The goal is to make the execution model understandable while exploring how it could support AI pipelines and other long-running jobs.
current boundary
durlo is pre-release and intentionally exploratory. It is not presented as a production-ready service yet; the interesting part is the execution model and the trade-offs behind making work durable.