Home

Musubi Wiki

Welcome. This wiki is the user-facing guide to running and integrating with Musubi. It complements — but doesn't replace — the architecture vault at docs/Musubi/ in the repo.

If you're trying to:

... do this	Start here
understand what Musubi is	Architecture Overview
get something running locally	Getting Started
deploy to a real host	Deployment Guide (draft — coming soon)
integrate an agent via MCP / LiveKit / OpenClaw	Adapters (draft — coming soon)
verify a published image signature	Supply Chain & Verification (draft — coming soon)
understand how the lifecycle engine decides what to keep	The Lifecycle (draft — coming soon)
write your own adapter against the SDK	SDK Guide (draft — coming soon)

Quick links

• Latest release
• Architecture vault (source of truth)
• Discussions — design questions, architecture musings, "how do I…"
• Issue tracker — bugs + feature requests

What Musubi is, in one paragraph

A memory server for a small fleet of AI agents. Episodic captures flow in, get scored and matured, cluster into synthesized concepts, and — when a concept survives reinforcement + age + contradiction gates — get promoted into a human-reviewable Obsidian vault. There's a lifecycle engine that runs these transitions on cron. There's a hybrid retrieval layer (dense + sparse + rerank) sitting on Qdrant. There's a thoughts plane for agent ↔ agent pub/sub. It's all one Python service; everything downstream (MCP, LiveKit, the CLI) is an adapter.

Contributing to the wiki

This wiki is editable by the maintainer. If you'd like to contribute a guide, please:

1. Open a Discussion proposing the page topic + outline.
2. Once green-lit, you can either PR a markdown file against the repo (fastest for me to review) or the maintainer will port your content into the wiki.

The wiki is not PR-reviewed the way the main repo is, so direct-commit access is limited. Canonical design content stays in the repo's docs/Musubi/ vault; the wiki is for using Musubi, not designing it.