// Agent profile
AgentOS by Rivet is an open-source operating system for AI agents that runs as an in-process lightweight virtual machine. It provides isolated POSIX-compliant environments with virtual filesystems, process tables, pipes, PTYs, and network stacks — all executing within the host process kernel with no container or VM bootstrapping required. Cold starts clock in at ~6ms (92x faster than traditional sandboxes), memory footprint sits at ~131MB per agent, and cost runs 6x cheaper than alternatives. AgentOS supports Claude Code, Codex, OpenCode, Pi, and Amp through a unified API with automatic conversation persistence, session management, and agent-to-agent delegation.
AgentOS is a lightweight operating system for AI agents built by Rivet. Rather than spinning up full containers or cloud sandboxes, it runs an in-process virtual machine that provides each agent with its own isolated Linux environment — complete with a virtual filesystem, process table, networking stack, and PTY support. The result is sandbox-grade isolation at a fraction of the latency and cost. It ships as an npm package, meaning you can embed agent execution directly into your Node.js backend without provisioning infrastructure.
The core value proposition is speed and cost. AgentOS achieves a p50 cold start of 4.8ms compared to seconds for container-based sandboxes, uses roughly 131MB of memory per coding agent (8x smaller than alternatives), and costs approximately $0.0000032/s on AWS ARM instances. Security follows a deny-by-default model — agents cannot access the filesystem, network, or spawn processes unless explicitly permitted. The unified API means you can swap between Claude Code, Codex, OpenCode, Pi, or Amp without changing your orchestration code. Built-in features include cron scheduling, webhook support, universal transcript formatting for debugging, and automatic conversation persistence with replay.
AgentOS fits anywhere you need to run coding agents programmatically at scale. SaaS platforms can offer agent-powered features without managing sandbox infrastructure. CI/CD pipelines can spawn isolated agent environments for code generation or review tasks with millisecond startup times. Multi-agent orchestration becomes practical when each agent runs in its own lightweight VM with built-in delegation and durable workflow chains. The external storage mounting (S3, Google Drive, host directories) makes it straightforward to give agents access to project files without exposing the host system.
AgentOS requires Node.js as the host runtime — it is an npm package, not a standalone binary. The isolation model uses WebAssembly and V8 isolates rather than hardware virtualization, which provides strong sandboxing but differs from traditional VM security boundaries. Agent support depends on registry-based command packages, so very new or custom agent CLIs may need packaging work. Licensed under Apache 2.0.
AgentOS targets platform engineers and backend developers who need to run coding agents in production without the latency, cost, or operational burden of traditional container sandboxes. If you are building an AI-powered development platform, a multi-agent orchestration system, or any product that programmatically spawns coding agents, AgentOS provides the runtime layer that makes that economically viable at scale.
Builder.io's open-source framework for building agent-native applications — shared actions, SQL-backed state, identity, tools, skills, jobs, observability, and UI surfaces that all work together.
CLI orchestration tool that installs and configures platform connectors for 16+ social and web sources — Twitter/X, Reddit, YouTube, GitHub, Bilibili, and more. Zero API fees.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.