Rivet is a specialized infrastructure designed for stateful, long-running "Actors" intended for AI agents, collaborative applications, and durable workflows. Unlike traditional infrastructure, Rivet Actors utilize in-memory state co-located with compute, achieving near-instant 0ms read latency by leveraging local SQLite or KV storage. The system is engineered for high efficiency and scalability, featuring a ~20ms cold start time and an extremely low memory footprint of approximately 0.6KB per instance, significantly outperforming Kubernetes pods (~6s start, ~50MB memory) and Virtual Machines (~30s start, ~512MB memory). Rivet supports advanced primitives including WebSockets for real-time bidirectional streaming, durable message queues, automated retries for multi-step workflows, and built-in scheduling. Architecturally, the platform offers a global edge network to deploy actors near users and can scale to zero to eliminate idle costs. Developers can use RivetKit for local development, host via a Rust-based engine using Postgres or FoundationDB, or utilize a fully managed cloud service. The ecosystem is built on an Apache 2.0 licensed, open-source foundation, ensuring infrastructure portability across various runtimes like Node.js, Bun, and Deno.