agentOS

agentOS is a lightweight virtual operating system designed specifically for AI agents. Instead of running each agent in a heavy sandbox or VM, you install agentOS as a library (npm package) and embed it directly into your existing backend. Each agent gets its own isolated environment with execution, filesystem, orchestration, and app deployment capabilities — all running in-process with minimal overhead. The result is cold starts in under 5 milliseconds, memory usage around 22 MB per instance, and costs that are 254 times cheaper than traditional sandbox providers.

What makes agentOS stand out is its architecture. It uses WebAssembly and V8 isolates — the same technology behind Chrome and Cloudflare Workers — to run multiple languages (Node.js, Python, Bash) inside a single compact runtime. Agents get a persistent POSIX filesystem that can mount S3, Google Drive, or host directories as normal paths. The orchestration layer lets you compose workflows, loops, multiplayer sessions, and agent-to-agent communication using ordinary JavaScript code. You can also deploy AI-generated applications for each user, all while keeping credentials safe through scoped access bindings.

agentOS is built for developers who are building production AI agent systems. If you're tired of managing heavy sandbox infrastructure or paying per-second VM costs, this library gives you a much lighter alternative. It works with popular agent frameworks like Pi, Claude Code, Codex, and OpenCode, and integrates with Rivet, Eve, and Flue. Use it to run agent code securely, orchestrate multi-agent workflows, or deploy custom apps — all from within your existing backend process.