graphify

Graphify Skill

Graphify turns a folder of code, docs, images, and videos into a queryable knowledge graph. After running graphify ., the generated GRAPH_REPORT.md is referenced by the assistant when you ask it to explore the codebase. Use it to navigate unfamiliar codebases, trace architectural intent, and share context efficiently.

Credential note: Semantic extraction (docs, PDFs, images) calls an external LLM using your own API key (ANTHROPIC_API_KEY or OPENAI_API_KEY). AST code extraction and Whisper transcription run fully locally with no API key required.

> Token efficiency: Reading GRAPH_REPORT.md is ~71.5× cheaper than reading raw source files. Always check it first.

Installation

Minimum install

pip install graphify

Recommended install with all extras

pip install "graphify[pdf,video,watch,svg]"

| Extra | Adds support for | |-------|-----------------| | pdf | PDF papers and documents | | video | Video/audio transcription via faster-whisper | | watch | --watch file monitoring | | svg | --svg export | | mcp | Model Context Protocol server | | neo4j | Neo4j graph database integration | | office | Word/Excel/PowerPoint documents |

Register with OpenClaw

graphify claw install

This copies the skill into OpenClaw's global skill directory and writes an AGENTS.md to the project root so the graph is consulted on every tool call.

Register with other platforms

graphify claude install     # Claude Code (CLAUDE.md + PreToolUse hook)
graphify cursor install     # Cursor (.cursorrules)
graphify codex install      # Codex (AGENTS.md)
graphify copilot install    # GitHub Copilot CLI
graphify gemini install     # Gemini CLI
graphify aider install      # Aider

Git post-commit hook (optional, local project only)

graphify hook install

Building a Knowledge Graph

First run

graphify .

Graphify runs a three-pass pipeline: 1. AST extraction — deterministic tree-sitter parsing of all code files (no LLM calls) 2. Transcription — local Whisper processing of any video/audio 3. Semantic extraction — parallel LLM calls (using your configured API key) analyze docs, papers, and images

Incremental update (changed files only)

graphify . --update

Watch mode (auto-sync as files change)

graphify . --watch

Deep mode (aggressive inferred-edge extraction)

graphify . --mode deep

Preserve edge directionality

graphify . --directed

Re-cluster without re-extracting

graphify . --cluster-only

Skip HTML visualization (faster CI runs)

graphify . --no-viz

Output Artifacts

All artifacts land in graphify-out/:

| File | Purpose | |------|---------| | GRAPH_REPORT.md | Read this first. God nodes, community structure, surprising connections, suggested questions. | | graph.html | Interactive browser visualization — open for human review. | | graph.json | Raw graph data for programmatic querying via CLI or script. | | cache/ | SHA-256 incremental cache — commit everything *except* this directory. |

Additional export formats

graphify . --svg        # SVG visualization
graphify . --graphml    # Gephi / yEd compatible export
graphify . --wiki       # Wikipedia-style article per node

Querying the Graph

# Natural-language semantic search
graphify query "where is authentication handled?"
Trace a specific path (DFS traversal)
graphify query "how does the request reach the database?" --dfs
Shortest path between two nodes
graphify path "AuthMiddleware" "PostgresAdapter"
Plain-language explanation of a node
graphify explain "UserSessionManager"

Adding External Content

graphify add      # Fetch and index a research paper
graphify add      # Fetch and index a tweet
graphify add      # Download and transcribe video/audio

After adding, run graphify . --update to integrate the new nodes into the graph.

Ignoring Files

Create .graphifyignore in the project root (same syntax as .gitignore):

node_modules/
dist/
build/
.next/
vendor/
*.generated.*
*.min.js
*.lock
graphify-out/cache/

See templates/graphifyignore.txt in this skill for a comprehensive starter.

Relationship Types

| Tag | Meaning | |-----|---------| | EXTRACTED | Found directly in source (AST, explicit reference) | | INFERRED | Reasonable inference; includes confidence score 0.0–1.0 | | AMBIGUOUS | Low-confidence; flagged for manual review |

Use --mode deep to maximize INFERRED coverage. Filter AMBIGUOUS edges when precision matters.

Best Practices for OpenClaw / Claude Code

1. Always read the report before exploring

Read graphify-out/GRAPH_REPORT.md

2. Map communities to features

3. Use graphify query before grep

4. Use graphify path for impact analysis

5. Use --update aggressively

6. Team workflows

7. Multimodal context

Supported Languages

Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, Ruby, C#, Kotlin, Scala, PHP, Swift, Lua, Zig, PowerShell, Elixir, Objective-C, Julia, Verilog, SystemVerilog, Vue, Svelte, Dart

Plus: Markdown, MDX, HTML, plain text, RST, PDF, PNG/JPG/WebP/GIF images, MP4/MOV/MKV/WebM/MP3/WAV/M4A/OGG media.

Troubleshooting

| Symptom | Fix | |---------|-----| | graphify: command not found | Run pip install graphify and ensure pip's bin directory is on PATH | | Graph is stale after edits | Run graphify . --update | | Missing nodes for a language | Confirm tree-sitter grammar is installed; run graphify . --update | | Video transcription slow | Expected — Whisper runs locally. Add GPU acceleration or use --no-viz to skip unrelated steps | | AMBIGUOUS edges dominating | Switch from --mode deep to default mode, or filter by confidence > 0.7 in graph.json | | AGENTS.md not picked up | Re-run graphify claw install from the project root |

| Symptom | Fix | |---------|-----| | graphify: command not found | Run pip install graphify and ensure pip's bin directory is on PATH | | Graph is stale after edits | Run graphify . --update | | Missing nodes for a language | Confirm tree-sitter grammar is installed; run graphify . --update | | Video transcription slow | Expected — Whisper runs locally. Add GPU acceleration or use --no-viz to skip unrelated steps | | AMBIGUOUS edges dominating | Switch from --mode deep to default mode, or filter by confidence > 0.7 in graph.json | | AGENTS.md not picked up | Re-run graphify claw install from the project root |