graphite-graph 0.1.2

Mavis-native knowledge graph skill. Cross-corpus (code, ops, notes, images) with PII-safe extraction and Mavis memory wiring.

graphite-graph

Mavis-native knowledge graph skill. Cross-corpus (code, ops, notes), privacy-aware, and wired into Mavis memory.

The binary is gph (not graphite) to avoid colliding with the unrelated graphite PyPI package (text graph renderer, v0.71).

What this is

A Mavis-shaped fork of graphify, with three deliberate additions:

	graphify	graphite-graph
Multi-corpus (code + ops + notes)	single-folder	unified corpus tag on every node
PII safety	none	.graphite-redact config + per-node pii_level
Memory wiring	memory folder (graphify #1441, in proposal)	writes back to AGENTS.md / MEMORY.md
Provenance	EXTRACTED / INFERRED / AMBIGUOUS	same — kept verbatim
Platform bet	cross-platform (Claude, Codex, OpenCode, ...)	Mavis + Python
Clustering	Leiden via graspologic	Leiden via leidenalg (Python-only)
CLI binary	graphify	gph (no collision)

Install

# base install — gets you build/query/path/explain/redact/inspect
pip install graphite-graph

# with clustering (Leiden)
pip install graphite-graph[cluster]

# with MCP server (so Mavis agents can query the graph live)
pip install graphite-graph[mcp]

# all of the above
pip install graphite-graph[cluster,mcp]

Use

gph build ./corpus           # build graph
gph query "tenant"           # ask the graph
gph path "AuthService" "DB"  # shortest path
gph explain "RateLimiter"    # explain a node
gph serve ./graph.json       # expose as MCP server

What you get

• graphite-out/GRAPH_REPORT.md — readable report (god nodes, surprising connections, communities, suggested questions)
• graphite-out/graph.json — persistent, queryable graph
• graphite-out/graph.html — interactive viewer (M2 stub; full viz landing soon)

Status

M3 complete. TestPyPI: https://test.pypi.org/project/graphite-graph/0.1.0/. Production release pending after this preview.

Milestone	Status
M0 — pipeline shape (code-only corpus)	done, 74 tests
M1 — multi-extractor (text, ops, image)	done, 101 tests
M2 — cluster + analyze + query + MCP	done, 128 tests
M3 — PII redaction + memory wiring + SKILL.md	done, 144 tests