graphify

/graphify

Turn any folder of files into a navigable knowledge graph with community detection, an honest audit trail, and three outputs: interactive HTML, GraphRAG-ready JSON, and a plain-language GRAPH_REPORT.md.

Three things it does that Claude alone cannot:

1. Persistent graph — relationships are stored in graphify-out/graph.json and survive across sessions. Ask questions weeks later without re-reading everything.
2. Honest audit trail — every edge is tagged EXTRACTED, INFERRED, or AMBIGUOUS. You know what was found vs invented.
3. Cross-document surprise — community detection finds connections between concepts in different files you would never think to ask about directly.

Command surface

/graphify [<path>|<github-url> ...]   # build a graph (default path: current directory)
/graphify query "<question>"          # traverse the graph for an answer (--dfs, --budget N)
/graphify path "A" "B"                # shortest path between two concepts
/graphify explain "Concept"           # plain-language explanation of one node
/graphify add <url>                   # fetch a URL into ./raw, then update the graph

Build flags (combine freely): --mode deep, --update, --directed, --cluster-only, --no-viz, --svg, --graphml, --neo4j[-push <uri>], --mcp, --watch, --wiki, --obsidian[-dir <path>], --branch <b>, --whisper-model <name>. Multiple URLs build each repo and merge into one cross-repo graph.

Route by command

Pick the matching reference, then follow its steps in order. Do not skip steps.

• Building a graph (any /graphify <path> or <github-url>, with or without build flags) → reference/build-pipeline.md (Steps 0–9: clone, install, detect, transcribe, extract, build, cluster, label, export, report).
• Querying an existing graph (query, path, explain) → reference/query-and-navigate.md.
• Keeping a graph current (--update, --cluster-only, add, --watch, git hook, CLAUDE.md integration) → reference/update-and-maintain.md.

The interpreter resolved in build Step 1 is written to graphify-out/.graphify_python; every later bash block uses $(cat graphify-out/.graphify_python) instead of python3. Query and maintenance subcommands re-resolve it first if graphify-out/ was deleted (guard in the reference files).

Honesty Rules (apply to every branch)

• Never invent an edge. If unsure, use AMBIGUOUS.
• Never skip the corpus-size check warning (Step 2: >2M words or >200 files).
• Always show token cost in the report.
• Never hide cohesion scores behind symbols — show the raw number.
• Never run HTML viz on a graph with more than 5,000 nodes without warning the user.