growmos

A living knowledge graph that grows with your repo — shared, provenance-carrying memory for humans and AI agents. Agent-native. Zero dependencies. MIT.

pip install growmos
cd your-repo && growmos init      # detects Claude Code / Codex / Cursor / Gemini and wires them
growmos view                          # open the interactive explorer

Live demo → growmos's own graph

The repository dogfoods itself: 47 nodes, 56 edges, hub profiles, edge provenance. Click any node.

Live demo → the Apollo corpus

The Anthropic playbook's six-document example: "Edwin Aldrin" → "Buzz Aldrin", one connected component, density 1.58.

GitHub

Source, docs, methodology, issues. MIT.

PyPI

pip install growmos · Python ≥ 3.9 · no dependencies.
growmos view screenshot

How it works

Docs, ADRs, READMEs and sessions are extracted into typed entities and short-verb-phrase relations, resolved into canonical nodes ("Edwin Aldrin" → "Buzz Aldrin"), assembled with provenance and corroboration counts, and queried as k-hop subgraphs the agent must cite. The CLI does the deterministic work; the judgment work is handed to whatever agent you already run — Claude Code, Codex, Grok, Cursor, Gemini — as task packets. No API key needed. An evaluation loop (gold sets, F1, a 10-item readiness checklist) keeps it honest, automatically.

Read the methodology: METHODOLOGY.md.