Skip to content
Homeostat

Homeostat

A team knowledge base for coding agents that keeps itself current. Homeostat keeps a team’s conventions, decisions and system documents, gives every agent the part that matters for its task, and keeps them right: it learns from code review, from what developers teach their agents and from the code, scores every change, and retires what goes stale.

The knowledge lives as Markdown in a git repo you own (one folder per code repo when a team owns several, with rules grouped into topic files), changes only by pull request, and reaches every agent: Claude Code (plugin and hooks), any MCP client, and Cursor, Copilot and AGENTS.md through exported rules files.

It runs on what you already have: your git host, your CI, your developers’ sessions and your model credential. There is no Homeostat service, account or telemetry (data handling).

How it works

Section titled “How it works”
  1. Start. homeostat bootstrap builds a first base for each repo from its checkout: its existing docs, plus documents on its architecture, data and infrastructure.
  2. Capture. People and agents add lessons, which apply to their author’s sessions at once. A daily run in your CI also reads merged pull requests and drafts a rule wherever a reviewer’s comment led to a code change. It then packs each developer’s lessons into one PR, merges the confident ones, and sends only the uncertain ones to a reviewer.
  3. Govern. Every lesson gets a confidence score, which weighs the role of whoever vouches for it. The score merges it, declines it, or sends it to the team’s named reviewers (else the team that owns its area). Nothing with a safety finding merges on its own. A weekly deep dive merges duplicates, flags contradictions and retires stale rules.
  4. Deliver. Each session gets its repo’s rules and the document sections for the files it touches, within a token budget, with the rest listed by topic for the agent to fetch.

Guide

Section titled “Guide”
  • Quickstart: a working knowledge base in ten minutes
  • Setup: homeostat setup, the guided checklist, and what each optional step gives you
  • How a team uses it: the admin’s setup, each developer’s join, how every clone stays current, the daily run and the review queue
  • Rules and documents: the file formats and layouts
  • A first base: homeostat bootstrap, from your repos’ docs, schemas and deploy files
  • Moving to 0.2: repo folders and topic files for an existing KB
  • Configuration: every key in homeostat.yaml
  • The agents: the daily run, the review queue, sync, deep dive, bootstrap and seed, the digest, and model runtimes
  • Using it from your tools: the Claude Code plugin, MCP, and exports
  • Dashboard
  • Measuring selection: homeostat eval, cases for the rules each session should get
  • Before and after: homeostat bench, the same tasks through Claude Code with the plugin off and on

Trust

Section titled “Trust”

Project

Section titled “Project”

Architecture · Design and execution plan (historical) · Contributing · GitHub · Apache-2.0