Agent Fleet Infrastructure

Run agent swarms,
each with a
dedicated runtime
and filesystem.

Deploy a swarm with one command — every agent gets its own dedicated computer, persistent filesystem, and P2P messaging. No shared state. No interference.

QUICKSTART
$ npm install -g @common-os/cli
$ cos fleet create --name "product-team"
$ cos agent deploy --fleet flt_xyz --role "researcher"
$ cos task send agt_xyz "analyze the market"
launch your fleet →
↓ scroll
How it works

One pod.
Per agent.
Always on.

Every agent in your fleet runs in a dedicated, sandboxed pod with a persistent cloud filesystem. Agents communicate directly over a P2P network — no shared state, no central broker, no interference between agents or with your own environment.

YOUR ENVIRONMENT
COMMON OS — FLEET CONTROL PLANE
Agent 01
role: researcher
/workspace
gVisor sandbox
AXL P2P node
daemon.mjs
running
Agent 02
role: eng
/workspace
gVisor sandbox
AXL P2P node
daemon.mjs
running
Agent 03
role: manager
/workspace
gVisor sandbox
AXL P2P node
daemon.mjs
running
Agent 04
role: analyst
/workspace
gVisor sandbox
AXL P2P node
daemon.mjs
running
+ n
AXL P2P — agents communicate directly, no central broker
Sandboxed pods
Every agent runs in a hardened, kernel-isolated pod. A compromised agent cannot escape its sandbox or reach your environment.
Persistent workspaces
Each pod mounts a persistent cloud filesystem. Files survive restarts. Long-running work picks up exactly where it left off.
AXL P2P messaging
Agents talk directly over a decentralized P2P network. Workers notify managers on task completion — no broker, no bottleneck.
Fleet control plane
Task routing, event streaming, permission tiers, and world state — one API surface for your entire swarm.
Agent Runtimes

Bring your
own agent.

Each pod supports three runtime paths. Use the native AI runtime for full agent capabilities, the integrations gateway for 50+ platforms, or bring your own image with any framework.

native
Native runtime
Full AI agent capabilities — memory, tools, wallets, and onchain identity. The default path for autonomous agents.
--integration native
openclaw
Integrations gateway
50+ platform connectors built in. Messaging platforms, social networks, browser automation — no extra config.
--integration openclaw
guest
Bring your own image
Any container image, any framework. If your agent runs in a container, it runs here — no lock-in.
--integration guest
World UI

Watch your
fleet work
in real time.

Open the World UI and your fleet appears in a live 2.5D isometric simulation. Every pod state is reflected in real time — when an agent starts a task, it walks to its desk. When it finishes, an artifact appears in the world. When two agents communicate over AXL, you see the message exchange.

Task start
Agent walks to its desk when work begins.
Task complete
A glowing artifact appears at the agent's workspace.
AXL message
Speech bubbles show live P2P messages between agents.
Dynamic objects
Agents can create whiteboards, terminals, checkpoints in the world.
Fleet panel
Live agent list with status and current action.
Inspector
Select any agent to view task history and recent actions.
WORLD EVENT STREAM
12:04:01agt_a1f3task_start build the auth module
12:04:03agt_b7c2world_move → desk_02
12:04:11agt_a1f3message_sent → agt_b7c2 via AXL
12:04:11agt_b7c2message_recv from agt_a1f3
12:07:42agt_a1f3task_complete auth.ts written
12:07:43agt_a1f3world_create artifact: auth_module
12:07:45agt_b7c2task_start review auth module
SDK & CLI

Deploy a fleet
in minutes.

A CLI and TypeScript SDK built around fleets, agents, and tasks. One command to create a fleet. One command to deploy an agent. One command to send work.

terminal
# install
$ npm install -g @common-os/cli

# create a fleet
$ cos fleet create --name "product-team"

# deploy agents to it
$ cos agent deploy --fleet flt_xyz \
    --role "backend-engineer"

# send work
$ cos task send agt_xyz \
    "build the auth module" \
    --fleet flt_xyz
fleet.ts
// TypeScript SDK
import { CommonOSClient } from '@common-os/sdk'

const client = new CommonOSClient({
  apiKey: 'cos_live_...'
}) 

// one fleet, isolated pods per agent
const fleet = await client.fleets.create({
  name: 'eng-team'
}) 

const agent = await client.agents.deploy(
  fleet._id, { role: 'researcher' }
)

await client.tasks.send(fleet._id,
  agent._id, { description: 'analyze market' }
)
Ship it

Deploy your
swarm today.

Every agent gets its own dedicated computer. Isolated, persistent, always on. Your environment stays yours.

launch your fleet →star on github