actors

Why we included this project

Keeping state alive across requests is usually the hard part of building AI agents, realtime collaboration, or durable multi-step workflows, not the model call itself. Rivet Actors handles that with a single runtime primitive: in-memory state lives next to compute, persists automatically, and hibernates when idle, so you keep serverless economics. You get one long-running process per agent, session, or user, with built-in queues, scheduling, and WebSocket broadcast, which removes much of the glue code teams otherwise write around stateless functions and external databases. It suits Node.js and Bun backends that want an actor model without cloud vendor lock-in, since it runs on Kubernetes, AWS, a VPS, or the hosted Rivet Cloud. The main tradeoff to weigh is operational maturity: running a stateful runtime yourself versus paying for the managed option.