rivet-dev/rivet

Rivet Actors are the primitive for stateful workloads. Built for AI agents, collaborative apps, and durable execution.

What it solves

Rivet provides a high-density, scalable orchestration layer specifically designed for agentic workloads. It solves the problem of managing thousands or millions of stateful, long-lived AI agents that require durable memory, independent storage, and fast startup times without the overhead of traditional container orchestration like Kubernetes.

How it works

Rivet uses a "virtual actor" pattern where agents are treated as Actors—lightweight, durable processes. Unlike pods, Actors live inside existing worker processes (Node.js, Bun, or Rust) rather than in separate VMs or containers.

Key technical mechanisms include:

  • Durable State: Each Actor has its own SQLite database or POSIX filesystem, tiered to S3 for long-term storage.
  • Hibernation: Idle Actors release resources and hibernate, waking up in approximately 12ms when a new request arrives.
  • Control Plane: A single Rust-based control plane handles the scheduling, routing, and persistence of Actors across a cluster.
  • Connectivity: Actors can be addressed directly via HTTP or WebSockets, allowing for real-time event broadcasting to clients.

Who it’s for

  • AI Agent Developers: Those building coding agents, personal assistants, or company-specific agents that need long-term memory and durable state.
  • Agent App Builders: Developers creating platforms where every user gets their own dedicated, scalable backend (Actor).
  • Real-time Application Developers: Those building collaborative tools (e.g., rooms or channels) that the same architecture can support.

Highlights

  • Extreme Efficiency: 12ms cold starts and only 72KB of memory per instance.
  • Built-in Persistence: Integrated SQLite and filesystem support per Actor.
  • Flexible Deployment: Open-source and self-hostable (Apache 2.0), with options for managed cloud or BYOC (Bring Your Own Cloud).
  • Developer Experience: Typed SDKs for backend and first-party React hooks for the frontend.
  • Fault Tolerance: Automatic rescheduling of Actors to other workers if a worker fails, maintaining state integrity.

Related

  • Project
    cloudflare/agents

    Cloudflare Agents is a TypeScript SDK that turns Cloudflare Durable Objects into persistent, stateful “agents” with built‑in real‑time sync, scheduling, AI chat, tool execution, voice, email, payments, and more. It provides decorators for RPC, React hooks for the frontend, and a suite of packages for chat, code‑generation, and dynamic Workers. Ideal for per‑user assistants, collaborative apps, voice bots, and workflow engines.

  • Project
    rivet-dev/agents

    Agent infrastructure from Rivet that allows running AI agents as durable actors and providing secure sandboxes for coding agents.

  • Project
    letta-ai/letta

    Letta is an open‑source framework for building stateful AI agents with persistent memory. It provides a CLI, a self‑hosted server, channel integrations (Slack, Telegram, Discord), a TypeScript SDK, and optional cloud storage for memory, enabling agents that can learn and retain context over time.

  • Project
    rivet-dev/dynamic-apps

    Rivet Dynamic Apps is a framework for securely executing user‑provided request‑handler code in sandboxed V8 isolates, with versioned state stored in Rivet actors and a fast cache‑hit path for low‑latency responses. It targets AI/LLM workloads by allowing custom, updatable logic to run safely at the edge.

  • Project
    agent-substrate/substrate

    Agent Substrate is a Kubernetes‑native runtime that multiplexes millions of stateful “agent” sandboxes (actors) onto a small pool of worker pods, offering sub‑500 ms suspend/resume, full memory‑disk snapshots, and strong isolation via gVisor or micro‑VMs. It provides a control‑plane API, node supervisor, and Envoy‑based routing, enabling high‑density, low‑latency execution of AI agents (LangChain, Claude Code, etc.) and any stateful container workload.