An open-source computer for agents that you can run anywhere. Faster, cheaper, and less memory than sandboxes.

Works with
PiPi
Claude CodeClaude Code*
CodexCodex*
OpenCodeOpenCode*
AmpAmp*
*Coming Soon
Get started

Up and running in seconds.

A full agent — sessions, tools, and isolated code execution — in a few lines. It's just an npm package.

agents.ts
import { AgentOs } from "@agent-os/core";
 
const vm = await AgentOs.create();
 
// Create an agent session
const { sessionId } = await vm.createSession("pi");
 
// Stream events (tool calls, text output, etc.)
vm.onSessionEvent(sessionId, (event) => console.log(event));
 
// Send a prompt and wait for the response
await vm.prompt(sessionId, "Write a Python script that calculates pi");

A new operating system architecture.

agentOS gives every agent — and the code it runs — its own lightweight VM: a V8 isolate with WebAssembly, not a full sandbox, with many packed into a single process. It's a real operating system for agents: an in-process kernel hands each VM its own filesystem, processes, and network, exposes your backend functions as host tools the agent calls directly, and schedules and bounds every isolate on its own — the flexibility of Linux at a fraction of the overhead.

One OS for every agent.

The agent sits at the center. The OS brokers Tools & Resources over MCP and host tools, persists Session state, mounts a Sandbox for heavier code, and drives Orchestration — so any supported agent gets the same capabilities with no bespoke glue.

WebAssembly + V8 isolates

The same isolation that powers Chrome — high-performance virtualization with no specialized infrastructure.

Battle-tested

You're using it to view this page. Bring it to your agents — no VMs, no containers, no overhead.

Faster, lighter, cheaper than sandboxes.

No VM to boot, no container to pull, no full gigabyte reserved per idle agent. agentOS cold-starts in about 5 ms and packs far more work into the same memory.

Cold startshown ~6× slower
Containers — one process each
~0 ms
agentOS — one shared processWhat's measured: Time from requesting an execution to first code running.

Why the gap: agentOS runs agents in-process — V8 isolates and Wasm inside your host. No VM to boot, no network hop, no disk image. Sandboxes must boot an entire environment, allocate memory, and establish a network connection before code can run.

Sandbox baseline: E2B, the fastest mainstream sandbox provider as of March 30, 2026.

agentOS: Median of 10,000 runs (100 iterations x 100 samples) on Intel i7-12700KF.
92× faster~0 ms

Each container is its own process that boots on its own; agentOS packs every agent into one shared process that boots once. Real p50 cold start: ~5 ms vs ~440 ms (92× faster) — container boot shown slowed ~6×.

Memory Per Instance
0xsmaller
agentOSWhat's measured: Memory footprint added per concurrent execution.

Why the gap: In-process isolates share the host's memory. Each additional execution only adds its own heap and stack. Sandboxes allocate a dedicated environment with a minimum memory reservation, even if the code inside uses far less.

Sandbox baseline: Daytona, the cheapest mainstream sandbox provider as of March 30, 2026. Default sandbox: 1 vCPU + 1 GiB RAM.

agentOS: ~131 MB for a full Pi coding agent session with MCP servers and file system mounts.
~131 MB
Cheapest sandbox~1024 MB

Sandboxes reserve idle RAM per agent; agentOS isolates share the host.

~1024 MB
Sandbox
~0 MB
agentOS
Cost Per Execution-Second
0xcheaper

$0.0084/hr server ÷ 4 executions = $0.00000058/s per execution-second

agentOSWhat's measured: server price per second / concurrent executions per server

Why it's cheaper: Each execution uses ~131 MB instead of a ~1024 MB sandbox minimum. And you run on your own hardware, which is significantly cheaper than per-second sandbox billing.

Sandbox baseline: Daytona, the cheapest mainstream sandbox provider as of March 30, 2026. Default sandbox: 1 vCPU + 1 GiB RAM at $0.0504/vCPU-h + $0.0162/GiB-h.

agentOS: ~131 MB baseline per execution, assuming 70% utilization (industry-standard HPA scaling threshold). Select a hardware tier above to compare.
$0.00000000/s
Cheapest sandbox$0.000018/s

Assumes one agent per sandbox, needed for isolation.

One server — AWS ARM4 executions
One per sandbox1 execution
idle · reserved

~131 MB used of ~1 GB reserved — the rest is billed while idle.

Measured on Intel i7-12700KF. Cold start baseline: E2B, the fastest mainstream sandbox provider as of March 30, 2026. Cost baseline: Daytona, the cheapest mainstream sandbox provider as of March 30, 2026 (1 vCPU + 1 GiB default). Cost assumes 70% utilization on self-hosted hardware vs. per-second sandbox billing. Benchmark document

Meet your agent's new operating system.

Embed in your backend.

Your APIs. Your toolchains. No complex agent authentication needed. Just JavaScript functions or hooks.

Mount anything as a file system.

S3, SQLite, Google Drive, or the host file system. No per-agent credentials needed.

Agents think in files. agentOS lets you expose any storage backend as a familiar directory tree. The host handles credential scoping, so agents never see API keys or secrets.

Granular security.

Fully configurable network and file system security. Control rate limits, bandwidth limits, and file system permissions. Set precise CPU and memory limitations per agent.

Your laptop, your infra, or on-prem.

Rivet, Railway, Vercel, Kubernetes, and more. Deploy wherever your code already runs.

agentOS is just an npm package. No vendor lock-in, no special infrastructure. Your agents run in your stack, on your terms.

RivetRailwayVercelKubernetesECSLambdaGoogle Cloud Run

Agents that just work.

Every agent deserves a runtime that understands it.

Supports Claude Code, Codex, OpenCode, Amp, and more

Run any coding agent with a single unified API. Swap agents without changing your infrastructure.

Simple sessions API

Create, manage, and resume agent sessions with a few lines of code. State persists automatically.

Coming Soon

Embedded LLM metering

Track token usage, cost, and latency per agent. No per-agent API keys needed. The host handles credential scoping.

Universal transcript format

One transcript format across all agents. Powered by ACP. Compare, debug, and audit any session.

Automatic transcript persistence

Every conversation is saved. Replay sessions, audit behavior, and build on past context without extra code.

Infrastructure that disappears.

Deploy anywhere. Scale to anything. Forget about servers.

Runs on your infra

Managed hosting or self-hosted. Same API, same experience, your choice of where it runs.

Easy to deploy on prem

A single npm package. No Kubernetes operators, no sidecar containers. Just install and run.

Low overhead

No VMs to boot. No containers to pull. Start in milliseconds with minimal memory footprint.

Mount anything as a file system

S3, GitHub, databases. No per-agent credentials needed. The host handles access scoping.

Extend with a sandbox when needed

agentOS handles most tasks, but pairs seamlessly with sandboxes for heavier workloads.

Orchestration without complexity.

Coordinate agents, humans, and systems out of the box.

Authentication

Authenticate agent connections with your existing auth model. Validate credentials and attach user state on connect.

Webhooks

Receive external events and route them into agents with lightweight HTTP handlers and durable queues.

Multiplayer & Realtime

Multiple clients can observe and collaborate with the same agent environment in real time.

Agent-to-Agent

Let agents delegate work to other agents through host-defined tools and shared orchestration flows.

Workflows

Chain agent tasks into durable workflows with retries, branching, and resumable execution built in.

Queues

Serialize agent work with durable queues for backpressure, async processing, and ordered execution.

SQLite

Give agents access to a persistent SQLite database through host tools for structured state and queryable memory.

Security without compromise.

The same isolation technology trusted by browsers worldwide.

Restrict CPU and memory granularly

Set precise resource limits per agent. No runaway processes, no noisy neighbors.

Programmatic network control

Allow, deny, or proxy any outbound connection. Full control over what your agents can reach.

Custom authentication

Bring your own auth. API keys, OAuth, JWTs. Agents authenticate on your terms.

Isolated private network

Each agent runs in its own network namespace. No cross-talk between tenants.

Powered by WebAssembly and V8 isolates

The same sandboxing technology behind Google Chrome. Battle-tested at planet scale.

After
Before
Unix Operators
agentOS Operators

Left: Unix timesharing, UW-Madison, 1978. Right: "Data flock (digits)" by Philipp Schmitt, CC BY-SA 4.0

From humans to agents

The operating system is changing for the next generation of software operators.

Learn more