rivet-dev/rivet

Rivet is a runtime for building stateful workloads using actors as the primitive, designed for AI agents, collaborative applications, and durable execution.

Rivet addresses the challenge of managing long-lived, stateful services by providing actors—lightweight processes that maintain in-memory state with automatic persistence. Each actor can run indefinitely while idle, hibernating to zero cost when not in use. State lives co-located with compute for instant reads and writes, persisting to SQLite or a custom database. The platform includes built-in WebSockets for real-time communication, workflows with automatic retries, durable message queues, and scheduling capabilities like timers and cron jobs. Actors scale infinitely and deploy across a global edge network, allowing you to run one actor per agent, session, user, or tenant without managing infrastructure complexity.

Rivet suits teams building AI agents with persistent context and memory, real-time collaborative applications, multi-step workflows, or per-tenant systems where low-latency in-memory access matters. It works well for sandbox orchestration, chat systems, and any workload that benefits from long-lived state without the operational overhead of traditional infrastructure. The README compares actors favorably to Kubernetes pods and virtual machines on cold start time, memory overhead, and idle cost, and to Redis and Postgres on read latency for in-memory state access.

The project shows active development with regular updates to its documentation and changelog. The team maintains a public Discord community and social presence for user engagement. The codebase is written in Rust, indicating a focus on performance and reliability for the runtime layer.