Restate

Restate is a durable-execution runtime, founded by ex-Apache Flink engineers, that aims to be the simplest way to write reliable distributed code — including the kind of stateful, long-running AI agent workflows that crash, retry, and resume. You write normal TypeScript, Python, Java, or Go functions; Restate gives them durable state, exactly-once semantics, automatic retries, timers, and inbox-style messaging out of the box. If a function fails halfway through a 20-step agent loop, Restate replays it from the last durable step rather than restarting from scratch — the same value proposition as Temporal, but with a single self-contained binary and a developer experience that feels closer to writing a normal web service than learning a workflow engine. The Restate runtime is open source under the Business Source License (BSL); Restate Cloud is the managed offering in early access. For AI engineers, Restate is increasingly popular as the backbone for production agents: the agent loop, tool calls, and human approvals all become durable workflow steps that survive deploys, crashes, and rate limits. Best when you want durability without standing up Kafka, a workflow database, and a separate scheduler.