Self-hosting documentation
Operate Roost yourself
These pages cover the accountless open-source edition: one coordinator, a worker on every supported macOS or Linux machine you own, and a browser client — including Windows — that reaches all of it. They document installation, fleet and terminal mechanics, networking, the CLI, and the security model. The per-account managed implementation is qualified but not launched; production signup and the shared dashboard origin are inactive, and accounts can only be operator-created. To use the released product, start at Install.
Start
Get a coordinator, a worker, and your first browser session running.
Concepts
How the fleet, the terminal, agents, and the mobile client actually work.
- Fleet: coordinator, workers, keepersHow Roost splits work between a coordinator, outbound-only workers, and a keeper subprocess — plus per-machine metrics and fleet updates.
- Terminal fidelityCell-authoritative frames, raw versus status-fenced input, bounded recovery, a pinned VT core, predictive echo, mouse modes, and real hyperlinks.
- Agents and statusHow Roost observes agent state and sends occupant-fenced text to the same ordinary shell PTY without owning an agent process or conversation.
- Phones and tabletsThe phone and tablet client: PWA install, the on-screen key row, swipe gestures, why the soft keyboard never reflows the terminal, and iOS notifications.
Reference
Networking topologies, the full CLI surface, and the security model.
- NetworkingOne loopback coordinator listener, the front door you choose, and the paths that stay private.
- The roost CLIEvery roost subcommand, the release-matched agent skill, every roost api verb, and how roost api authorizes itself.
- Security modelHow Roost uses non-extractable device keys, pairing, revocation, audit, backups, and local-only telemetry.