I run coding agents on more than one machine. Claude Code on my laptop, another agent on a VPS, sometimes Pi on a Mac. They're all good at their jobs, but they can't talk to each other — and the options I found either wanted my API keys uploaded to someone's cloud, or amounted to SSH + tmux duct tape.
So I built Skep: a thin session bus that lets coding agents on different machines join one live session. It's open source (Apache-2.0).
The problem
- Agents are per-machine. Context, plans, and intent stay locked on the device where the agent runs.
- Most multi-agent setups centralize: your keys and your code flow through a third-party server.
- I wanted agents to share intent with each other, keys to stay on each device, and to keep running the local CLIs I already trust.
What Skep does
- One machine becomes the master; other machines join with a short code + host. No accounts, no cloud relay.
- Peers run your own local Claude Code / Codex / Pi. Skep doesn't wrap them or proxy your prompts through a server.
- Keys stay on each device. Skep never harvests or uploads API keys — there's no central server for them to go to.
- Install:
npm i -g @skepagent/skep(v0.1.15)
Why not a cloud platform?
Most "multi-agent collaboration" products hold your session state, your keys, and the billing meter on their servers. Convenient, but a bigger trust surface.
Skep inverts it: the session is just a bus between machines you already own, meant for trusted devices on your own network. Less magic, fewer places your secrets can leak.
Try it
- Site: https://skepagent.com
- Repo: https://github.com/Guxiaodong941005/skepAgent
npm i -g @skepagent/skep
I'm the author — happy to answer protocol / threat-model questions in the comments. If you run agents across machines, I'd love to hear how you'd use this.
Top comments (0)