Give every AI agent its own machine

Frontend, backend, devops — each agent on its own Linux VM with 200+ tools. One orchestrator coordinates them all. A dev team made of agents for $12/shell/mo.

Get started See how it works

Why agents need separate machines

The same reasons you don't run your whole team on one laptop.

Isolation

Each agent gets its own filesystem, processes, and network. No conflicts between a frontend build and a database migration running at the same time.

Parallelism

Five agents working simultaneously, not taking turns on one machine. Frontend builds while backend writes APIs while devops configures infrastructure.

Persistence

Every agent's session survives disconnects. Close your laptop, come back tomorrow — they're all still running. Check in from your phone.

How orchestration works

The orchestrator sends tasks, monitors progress, reads output, and relays information between agents. Each agent runs Claude Code on its own isolated VM. Every shell gets a dev preview URL so you can see what each agent built.

Set it up in 5 minutes

Create your shells

Sign up, pick a plan for each shell. Name them by role — frontend, backend, devops, whatever your project needs. Each boots in seconds.

Start Claude Code on each

SSH into each shell or use the web terminal. Run claude. Set your API key once — it persists. The built-in /orchestrate skill teaches Claude how to coordinate with other shells.

Send tasks from the orchestrator

From any shell (or your local machine with latch installed), send tasks to your agents and monitor their progress in real time.

Preview what they built

Each shell gets a public dev preview URL. Open it in a browser to see the frontend, test the API, check the logs — without SSH.

The orchestrator in action

orchestrator

latch shells list

frontend shell-plus running 4096MB 2vCPU

backend shell-plus running 4096MB 2vCPU

devops shell running 2048MB 1vCPU

latch shells send frontend "build a React dashboard with auth\r"

latch shells send backend "write a REST API in Go with JWT auth\r"

latch shells send devops "write Docker Compose for a React + Go + Postgres stack\r"

# check on the frontend agent...

latch shells screen frontend

Created 12 components. Running on :3000.

# read the backend's full conversation log...

latch shells exec backend "tail -5 ~/.claude/projects/*/claude.jsonl"

{"role":"assistant","content":"API complete. 8 endpoints. All tests passing."}

send tasks, monitor progress, read full conversation logs, preview running apps — all from one terminal.

Every shell comes ready

No setup, no installing dependencies, no waiting.

200+ tools

Go, Rust, Node, Python, Docker, Terraform, kubectl, every build tool and language runtime your agents could need.

AI agents pre-installed

Claude Code, Codex, Aider, OpenCode, OpenClaw. Set your API key and go. The /orchestrate skill is built in.

Dev preview URLs

Every shell gets a public URL. Start a dev server on port 3000 and it's instantly accessible from any browser. No tunnels, no ngrok.

Docker & databases

PostgreSQL, Redis, MariaDB — one command to start. Docker and docker-compose are native. Your agents can run real infrastructure.

Access from anywhere

SSH, web terminal with Mosh, mobile app. Check on your agents from your phone. Sessions persist indefinitely through latch.

Public IPv6

Every shell gets a public IPv6 address. Agents can serve traffic directly, run webhooks, or communicate with each other.

Pricing

Per shell. Add as many as you need. Cancel anytime.

flat rate, per agent

Shell+

$12/mo

or $120/yr

4 GB RAM, 2 vCPU, 25 GB disk. More headroom for Claude Code, large builds, and Docker workloads.

Get started

3 agents = $36/mo. 5 agents = $60/mo. Add as many as you need.