Coding agents that don't need your laptop.
Start a long refactor, close the lid and check back later. Your agent keeps working on its own machine, and your repos, logins and history are still there next time.
Works with Claude Code and Codex
Get a private Linux machine with Claude Code and Codex already installed and a persistent disk. Send it a prompt or a shell command from a script, your CI, a chatbot or your phone, and watch the output stream back live.
Up to $6/month. Priced by the hour. Asleep when you're not using it.
Why agent-32
Start a long refactor, close the lid and check back later. Your agent keeps working on its own machine, and your repos, logins and history are still there next time.
Everything you can do in the dashboard you can do over the HTTP API, and commands and agent runs also stream over a realtime WebSocket or gRPC. Use agent-32 as the engine behind your own bots, pipelines and tools.
Instances go to sleep after 30 minutes of inactivity and wake automatically the next time you run a command or agent, open a terminal or visit an exposed port. While asleep you pay only for storage.
What you get
Log in once with your own Claude or OpenAI account, or set an API key. Logins are kept on your volume.
Your files, repos, tools and agent sessions survive restarts and sleep.
A full terminal in the dashboard, with tmux sessions that keep running after you close the tab, until the instance goes idle and sleeps.
Send a prompt and get Claude's or Codex's events as they happen, plus the final answer and a session id so the next prompt continues the same conversation.
Run commands in the foreground, stream them, or start them as background jobs with completion callbacks.
Upload and download files, and manage environment variables such as GITHUB_TOKEN that every job sees.
Every instance gets its own https://<id>.a32.es. Expose a port for a dev server or web UI and it's served there, either privately (behind your login or API key) or publicly.
Create keys limited to a single instance for bots and integrations.
API
REST
One HTTP call from any script, cron job or CI step.
curl -X POST https://a32.es/v1/instances/$ID/agent \ -H "Authorization: Bearer $A32_KEY" \ -d '{"agent":"claude","prompt":"Fix the failing tests and open a PR"}'
WebSocket
Carries many commands and prompts at once, with all output streamed back in real time.
→ {"id":"1","type":"agent","prompt":"Add rate limiting to the API"} ← {"type":"event","ref":"1","event":{"type":"agent", ...}} ← {"type":"done","ref":"1","status":"exited","result":"Done: added ..."}
gRPC
With streaming and a two-way realtime session.
grpcurl -import-path proto -proto a32/v1/a32.proto \ -H "authorization: Bearer $A32_KEY" \ -d '{"instance_id":"i-…","command":"npm test"}' \ grpc.a32.es:443 a32.v1.AgentService/Exec
How it works
Pricing
Deducted as you go. No subscription, no commitment.
Up to 2 vCPU · 4 GB RAM · 10 GB disk
$6.00 / month if always awake (720 hours)
FAQ
Yes. agent-32 provides the machine. You log in to Claude Code or Codex with your own account or API key.
Running processes stop, but everything in your home directory stays. Running a command or agent, opening a terminal or visiting its exposed port wakes it again. Running jobs keep it awake.
By default, headless agent runs skip Claude's and Codex's permission checks, because the container is the sandbox. Set safe_mode on an API request to keep those checks, and Codex's sandbox, on.
Yes. Expose a port and it's served at https://<id>.a32.es, either privately behind your login or API key, or publicly. Public visits wake the instance and count as awake time.
Each instance runs in its own isolated container with its own persistent volume.