Every coding agent, in a browser tab
Claude Code, Codex, opencode and Antigravity install with their makers’ own commands and run unmodified on real Linux, with your outpost as their sandbox.
$ claude
❯ add rate limiting to /orders
Read src/orders.ts
Edited src/orders.ts +18
Ran npm test
● Done: 100 requests a minute per key. 14 passed.
Brewed for 11s
$ npm run dev
VITE v7 ready in 1.2 s
-> Local: http://localhost:5173/
-> Preview: k3x9-5173.0cloud.run
hmr update /src/orders.ts
One click from the Agents folder
Open the Agents folder and pick one. The Store runs the maker’s install command in a terminal you can watch, and the agent is on your PATH a moment later.
- Claude Code: curl -fsSL https://claude.ai/install.sh | bash
- Codex: sudo npm install -g @openai/codex
- opencode: curl -fsSL https://opencode.ai/install | bash
- Antigravity CLI: curl -fsSL https://antigravity.google/cli/install.sh | bash
Full-auto mode, safely
The outpost is the agent’s sandbox. It can install packages, delete files and rebuild the project without touching your laptop. Take a snapshot before a big task and roll back if you don’t like the result.
- Agents cannot reach your laptop’s files
- Snapshot before, roll back after
- Bubblewrap sandboxes work inside the outpost
Close the window, the agent keeps working
Agent windows reattach. Close one while the agent is busy, open it again later and you are back in the same session, with everything it printed.
- Sessions survive a reload
- Pick up a long task on your phone
- Every agent in its own tab or split
Questions
Do I need my own account with the agent’s maker?
Yes. Sign in to the agent the way you would on your laptop; 0cloud does not resell model access.
Are the agents modified?
No. Each one installs with its maker’s own command and runs as it would on an arm64 Linux machine.
Is it safe to use full-auto mode?
The agent works inside your outpost, a separate Linux machine in the tab. It cannot reach your computer’s own files.