Graphify
Knowledge graph CLI for coding agents. Turns a folder (code in 25 languages via tree-sitter, Markdown, PDFs, images, videos/audio transcribed locally) into a queryable graph. Output: interactive HTML + graph.json + Markdown report listing 'god nodes', communities and surprising connections. SHA256 cache: re-runs only process changed files. Works with Claude Code, Codex, Cursor, Gemini CLI, and others.
Strengths
- Local AST parsing (tree-sitter, 25 languages) for code, no tokens billed on that part
- Incremental SHA256 cache: a re-run only reprocesses changed files
- Queries beyond the viz: `graphify query "what connects X to Y?"` and `graphify path` for shortest path
- Typed edges (EXTRACTED / INFERRED with score / AMBIGUOUS), each relationship carries its confidence
- Multi-runtime (Claude Code, Codex, Cursor, Gemini CLI…), one tool across your whole stack
Limitations
- Python 3.10+ required, you need the environment (or go via pipx)
- Docs and images go through the provider API (your key, your tokens), only code is 100% local
- Marginal token gain on small corpora (a few files), the value shows at scale
- Video/audio transcription requires an extra (`pip install graphifyy[video]` + local Whisper)
Best for
- Onboarding onto a medium-to-large codebase, the agent queries the graph instead of re-reading 500 files
- Multi-source research: merging code, Markdown notes, PDF papers and conference videos into one graph
- Spotting surprising dependencies or tightly coupled modules ('god nodes' listed in the report)
- Preparing a refactor or migration by first visualizing the actual code topology