Hangar

hangar

**Personal agent operating system.** A self-hosted control plane for AI coding agents (Claude Code, Codex, and others), shipped as a Linux container that runs locally under OrbStack on a Mac and identically on any cloud Linux VM. Think `tmux + kubectl + OBS` reimagined for someone who lives inside multiple Claude sessions.

Status: v0.2.0 shipped Phase 2 MVP (Rust backend, SvelteKit dashboard, ntfy push, REST prompt API). Now migrating from the original optiplex host to a containerized deploy on the Mac mini ([ADR-0017](docs/decisions/0017-containerize-deployment.md)).

---

Why this exists

Running several Claude Code / Codex sessions in parallel on a remote box via tmux + SSH works but has rough edges:

• No visibility — you have to attach to know what each agent is doing
• No automation — prompting from phone / iOS Shortcut / cron means SSH gymnastics
• No notifications — Claude stalls waiting for a yes/no and you have no idea
• No memory — once a session is closed, its history is gone
• No intelligence — nothing parses what Claude is actually doing, it's all terminal bytes

hangar is the tool that sits above tmux (and later replaces it) to give you a real control plane: web UI, push notifications, REST API, session history, multi-agent orchestration.

**Primary goal:** a command-center dashboard where you can see every running agent at a glance, click in to any session, drive prompts from anywhere, and be alerted when any session needs attention.

---

Who this is for

• Solo developers running multi-session AI-agent workflows
• Anyone who has an always-on box (Mac mini, home server, cloud VM) and a laptop/phone they hop between
• People who want to build on top of a typed session/agent model rather than scrape ANSI terminal bytes

Not for: shared team installs, multi-tenant Saa