If you run AI coding sessions (Codex, agents, background jobs) for more than a few weeks, they start to sprawl. One session on the laptop, one on a home server, a couple inside Docker containers on a remote VM — and no single place that tells you what is running, where it lives, and how to hand it off.
TaskHandoff is an open-source control plane (Apache-2.0, TypeScript) that pulls those AI workspaces into one board. I've been using it to keep local and remote machines in the same view, and it's finally at a state worth sharing.
The problem
Once AI sessions spread across machines, you end up juggling:
- SSH tabs, container shells, and a half-dozen config files
- No shared view of "what's running where"
- Rebuilding the same container toolchain over and over
- Copy-pasting context between a chat app and the machine that actually does the work
What it does
- Multi-node management — local and remote machines in a single board
- Managed Docker workspaces — create, run, and restore workspaces as first-class objects
- Environment templates — snapshot a container's tools + config and reuse it elsewhere
- AI session center — real-time sessions over WebSocket, attach/detach at will
- Repository workflows — files, changes, branches, and git worktrees from the UI
- Scoped Git credentials — hand each workspace only the credentials it needs
- Chat integrations — route tasks from Telegram, DingTalk, WeChat, or Lark
- App catalog + mobile client — plus a Desktop app and a server, with an English/Chinese UI
Quick start
curl -fsSL https://github.com/edgestorage/task-handoff/releases/latest/download/install-server.sh | sudo sh
Ships as a Desktop client and a server (systemd on Debian/Ubuntu). Once the server is up, open the console and add your machines.
Why it exists
Most AI tooling optimizes the model. TaskHandoff optimizes the plumbing — the boring, error-prone part of running those models somewhere real, across more than one box, without losing track of your sessions.
Links
Apache-2.0. Stars, issues, and PRs are all welcome — and I'd love to hear how you're juggling multi-machine AI sessions today.
Top comments (1)
tr.ee/dev-to