ashik-manjunath/graphify

Purpose and entry points

Graphify is a Python 3.10+ CLI that converts a repository or mixed corpus (code, docs, media, workspace exports, schemas) into a queryable knowledge graph. Primary entry points are graphify/__main__.py for the graphify command and python -m graphify, and graphify/serve.py for MCP serving. It produces graphify-out/graph.json, GRAPH_REPORT.md, and optional browser/call-flow visualizations.

Orchestration and graph build

The build flow is coordinated by graphify/build.py: discover input, extract nodes and relationships, merge with existing state when requested, cluster the graph, and emit artifacts. graphify/extract.py is the headless extraction entry point; graphify/analyze.py and graphify/detect.py classify files and drive content-specific handling. The important boundary is local structural extraction versus semantic extraction that may invoke an LLM.

Local source extraction and resolution

Code is parsed locally, primarily through AST/language extraction, then normalized into graph entities and edges. graphify/extract.py owns broad extraction; graphify/symbol_resolution.py resolves cross-file/import symbol references; graphify/semantic_cleanup.py and graphify/dedup.py reconcile duplicate or ambiguous entities. This stage creates the canonical graph vocabulary before reporting, querying, or export.

Non-code and external ingestion

Specialized adapters add sources beyond repository code: MCP configurations (graphify/mcp_ingest.py), SCIP indexes (graphify/scip_ingest.py), PostgreSQL schema introspection (graphify/pg_introspect.py), Cargo workspaces (graphify/cargo_introspect.py), and Google Workspace shortcuts/content (graphify/google_workspace.py). Optional runtimes/services include LLM providers, Ollama, AWS Bedrock, Google Workspace gws, PostgreSQL, and graph targets such as Neo4j or FalkorDB.

Persistence, incremental updates, and graph composition

graphify/manifest.py tracks corpus state for portable/incremental rebuilds, while graphify/cache.py stores reusable extraction results. graphify/affected.py identifies impact from changes, and graphify/watch.py supports continuous synchronization. graphify/global_graph.py registers and traverses graphs across projects. These components separate raw extraction state from the generated graph artifact.

Enrichment, clustering, and deliverables

graphify/cluster.py detects graph communities and supports labeling; graphify/report.py generates architecture-oriented findings such as highly connected nodes, cross-cutting relationships, confidence, and suggested questions. graphify/export.py produces alternate formats and graph database exports, while graphify/callflow_html.py generates the call-flow architecture HTML. Visualization is downstream of graph construction and can be skipped without changing extraction.

Consumption and assistant integration

Graph consumers query graph artifacts through CLI-oriented analysis and retrieval in graphify/querylog.py, graphify/analyze.py, and the MCP server in graphify/serve.py, which supports stdio and HTTP with optional API-key protection. graphify/hooks.py installs Git hooks for rebuild/merge behavior. Assistant-specific skill and persistent-instruction assets under graphify/skills/ and graphify/always_on/ are generated/maintained by tools/skillgen/gen.py, enabling query-first behavior across coding-agent platforms.