Rhizome docs
A coding harness that doesn’t forget. Rhizome runs its own agent loop on any model, free ones included, and keeps what matters as plain Markdown in a vault you own. You can open, edit, diff and revert those notes like any other file.
What you get
Section titled “What you get”- Memory you can read. Save an answer to the vault with one click. It becomes a Markdown file. In a Git repository, every change is a commit you can undo. Your vault is the agent’s memory
- A replay of where the agent went. Mycelium draws the folder a Prime run worked in as a map, and lights up each file the agent searched, read or edited. Mycelium
- Free models that switch for you. With Taproot, Rhizome routes a chat across free hosted models. When a provider fails or runs out, it moves to the next one. Free models and routing
- Its own agent loop. Rhizome sends your message to a model, runs the tools the model asks for, and repeats until the answer is done. That native loop is the only chat engine (ADR-0187). Agents, engines and models
It also has what you’d expect from a harness and a notes app: sessions you can resume, permission modes for tools, your choice of model, a rich Markdown editor, PDF and image previews, a wiki graph, and Git built in.
Where to go next
Section titled “Where to go next”| If you want to… | Read |
|---|---|
| Decide whether Rhizome is for you | Why Rhizome |
| Try it | Install, then First launch |
| Move over from Claude Code, Hermes Agent, Cursor or Prime | Coming from another harness |
| Chat and keep what matters | Chat with the agent |
| Fix something | Chat doesn’t answer |
Origins
Section titled “Origins”Rhizome is AGPL-3.0-or-later. Mycelium uses Mindwalk (MIT) © 2026 Ricko Yu. Taproot builds on OmniRoute and freellmapi (both MIT). See Credits and licences.