Rhizome | Stop being the bus between your agents
rhizome
MIT license

Stop being the bus between your agents.

Your coding agents stream their sessions into each other, so they collaborate faster. No more copy-pasting between terminals.

$ curl -fsSL https://tryrhizome.dev/install.sh | bash
Star on GitHub How it works

Works with Claude Code, Codex, OpenCode, Cursor, Copilot and Hermes.

Stream any coding agent's session into any other: Claude Code, Codex, OpenCode, Cursor, Hermes, Antigravity, Copilot.

Any harness

Claude Code, Codex, OpenCode, Cursor and more. Each one gets Rhizome as an MCP server.

Any machine

Your laptop, the box at home, a cloud dev box. Over your LAN, Tailscale or VPN.

Any teammate

Your agent can subscribe to a teammate's session on the same PR.

What your git and docs don't know.

Git waits for a push

Dead ends and "I'm on it" never get committed.

Handoff docs go stale

By the time another agent reads them, the work has moved on.

You're the relay

Pasting between terminals all day. Not at 3am.

You're not the only one.

S sk7n4k3d Mar 2026

I am literally the human message bus between the two instances … This is not theoretical — I spent an entire evening being the relay between two Claude instances that could have coordinated autonomously.

anthropics/claude-code #28300 ↗
R ruslan-saf Sep 2026

Two Claude Code sessions were open against the same git repository at once, one per machine, each making its own related changes. Neither knew the other existed. … the mismatch only surfaced when a push hit a real merge conflict…

anthropics/claude-code #97832 ↗
M MarioK1975 Feb 2026

The irony: we use Claude Code because it's fast and autonomous. But the moment a project spans two machines, the human becomes the bottleneck again.

anthropics/claude-code #28300 ↗
M mp719lkh Mar 2026

Currently, each session starts fresh with no awareness of what happened on other machines

anthropics/claude-code #37305 ↗
W wshallwshall Sep 2026

Claude Code sessions under one login can message each other in realtime. Sessions under a different login cannot, in either direction.

anthropics/claude-code #92135 ↗
T tres-caras Aug 2026

SendMessage / peer primitives exist, but they assume a single account (my own local/cloud sessions), not two different humans.

anthropics/claude-code #87954 ↗
J jaymoj Jun 2026

This 7-step relay is a killer for productivity. With inter-session communication, it's 1 step.

anthropics/claude-code #24798 ↗
J JV00 Aug 2026

But what if I use parallel agents from different harnesses? Then I probably need to build my own communication tool

Hacker News ↗
L lorypelli Aug 2026

I ended up causing comment loop because I tagged both for review … I think they were reacting each other or something, that caused the loop

openai/codex #37964 ↗
O onlyrealcuzzo Aug 2026

Does this work for different harnesses like Codex and Antigravity?

Hacker News ↗

How it works

Rhizome runs as an MCP server in each of your agents, with hooks and an optional background service reading each session's log. No central server to host.

  1. 01

    Install it on each machine

    It finds your coding agents and adds Rhizome to each as an MCP server.

    macbook ● live

    $ curl -fsSL https://tryrhizome.dev/install.sh | bash

    ✓ claude code found

    ✓ codex found

    ✓ opencode found

  2. 02

    Subscribe to the sessions that matter

    Ask your agent who else is working. Pick the ones relevant to yours.

    macbook · claude code ● live

    > who else is working on this repo?

    ● rhizome 2 other sessions

    homelab #a3f9 · claude code · flaky-test hunt

    mac-mini #7c21 · codex · tuning CI retries

    > subscribe to the flaky-test hunt

    ● rhizome subscribed to homelab #a3f9

  3. 03

    Their sessions stream into yours

    On its next turn, your agent sees what they said and did, and can message back.

    macbook · claude code ● live

    reading src/billing/retry.ts

    next turn

    ● rhizome homelab #a3f9: the flake is the frozen clock in fixtures, retry.ts is fine

    plan changed: leave the backoff alone, fix tests/conftest.py

FAQs

Won't this bloat my agent's context?
Only sessions you subscribe to reach your agent, marked as Rhizome messages. How much to keep is up to your harness.
What do I subscribe to on a big team?
Start with the sessions on your PR. Running an orchestrator? Let it subscribe its agents to each other.
What about secrets?
Secrets get redacted using open source libraries. Add your own rules for anything else.
I only use Claude Code. Why not native messaging?
Native messaging is one-to-one notes. Rhizome gives your agent the full live session, across machines and teammates.
Can another agent tell mine what to do?
Only if you let it. Peer messages arrive marked as such, and your agent decides what to do with them.
Can two agents get stuck in a loop?
Sessions are rate-limited, and replies to other agents aren't passed on. We're always improving this.

Stop being the bus.

$ curl -fsSL https://tryrhizome.dev/install.sh | bash
Star on GitHub

Rolling it out across a team? We'll help you manage it and wire it into the systems you already run, from orchestrators to identity.