v0.1.0¶
Release Date: 2026-04-09
Initial release of Graphize - LLM-powered CLI for transforming Go codebases into queryable knowledge graphs.
Highlights¶
- LLM-powered tool to turn Go codebases into queryable knowledge graphs
- Two-step extraction: deterministic AST + optional LLM semantic analysis
- TOON export format with 98% token reduction vs JSON for AI agents
- MCP server for AI agent integration with 5 graph query tools
CLI Commands¶
| Command | Description |
|---|---|
graphize init |
Create graph database directory structure |
graphize add <repo> |
Track repositories with git commit hashes |
graphize status |
Show sources with staleness detection |
graphize analyze |
Extract AST graph from Go sources |
graphize query |
BFS/DFS traversal with depth control and edge filters |
graphize export html |
Cytoscape.js interactive visualization |
graphize export json |
Cytoscape JSON format |
graphize export toon |
Token-efficient agent output |
graphize summary |
Markdown summary for agents |
graphize report |
God nodes, communities, surprises analysis |
graphize enhance |
Prepare files for LLM semantic extraction |
graphize merge |
Merge LLM-extracted semantic edges |
graphize serve |
MCP server for AI agent integration |
AST Extraction¶
- Go AST extraction: packages, files, functions, methods, types, imports, calls
- SHA256-based per-file caching for fast incremental rebuilds
--no-cacheflag to bypass cache when needed- Cache hit/miss statistics reporting
Analysis Features¶
- God nodes: Most connected entities (excluding file-level hubs)
- Community detection: Louvain algorithm with cohesion scores
- Surprising connections: Cross-file, cross-community edges
- Isolated nodes: Potential documentation gaps
- Package statistics: Files, functions, types, imports per package
- Edge confidence: EXTRACTED, INFERRED, AMBIGUOUS breakdown
Semantic Extraction¶
LLM-discovered edge types:
inferred_depends- Implicit dependenciesrationale_for- Design rationale from commentssimilar_to- Semantic similarityimplements_pattern- Design pattern implementationsshared_concern- Cross-cutting concerns
MCP Server Tools¶
| Tool | Description |
|---|---|
query_graph |
BFS/DFS traversal from search terms |
get_node |
Node details by ID or label |
get_neighbors |
Adjacent nodes with edge info |
get_community |
List community members |
graph_summary |
Stats, god nodes, suggested questions |
Storage¶
Uses GraphFS for git-friendly storage:
.graphize/
├── manifest.json # Tracked sources
├── nodes/ # One file per node
├── edges/ # One file per edge
└── cache/ # Per-file extraction cache