A spare Linux box becomes your agents' server room

What it does Cloudroom core is a single Rust service that runs coding agents (Claude Code, Codex, Pi, and Cursor) on a Linux machine you control, and persists their chat history to PostgreSQL. It’s the backend for the Cloudroom desktop app, but the app is optional — the core exposes a plain HTTP API and works standalone. Agents authenticate with your own logins, so there’s no Cloudroom account and no metered middleman.

The interesting bit The security model is the actual design, not an afterthought: each agent runs as a separate unprivileged Linux user with no sudo, and can’t read the core’s token, database password, or history files. A “Command Guard” blocks a short list of catastrophic commands — a pragmatic admission that agents run without approval prompts, so isolation matters more than permission dialogs.

Key highlights

• One Rust service per machine; no Cloudroom account required
• Works with Codex, Claude Code, Pi, and Cursor, using your own model logins
• Chat history stored in PostgreSQL, local or remote over verified TLS
• API is token-authenticated and localhost-only unless you put it behind HTTPS
• Tested Linux binaries shipped per release; the core never runs database migrations itself

Caveats

• Explicitly pre-release software, per the README
• Agents run without approval prompts — the docs themselves say give each trusted user their own machine
• Setup targets Ubuntu 24.04; the desktop app is Apple Silicon only (Linux in alpha)

Verdict If you already have a VPS or a spare laptop and want your coding agents running somewhere that isn’t your laptop, this is a tidy, security-conscious way to do it. If you want a managed multi-tenant setup, wait — hosted Cloudroom is invite-only for now.