Introducing Graphify: a knowledge graph for your codebase, inside your coding assistant

Every AI coding assistant explores your codebase the same way: it greps. It searches for strings, opens the files that match, and tries to reconstruct the architecture in its head — every single session, from scratch. That works for small repos and falls apart the moment your question spans more than one hop: what actually calls this function, what breaks if this schema changes, why does the billing service import the auth client at all.

Today we're launching Graphify publicly: an open-source (MIT) skill for your AI coding assistant. Run /graphify . and it maps your codebase into a knowledge graph — real nodes, typed edges — that your assistant queries instead of grepping. It's already at roughly 77.6k stars on GitHub, and we're part of YC's Summer 2026 batch.

One command, three files

Run /graphify . in any repo and you get three local files. graph.html is an interactive map of your codebase you can actually explore. GRAPH_REPORT.md is the written brief: the god nodes everything depends on, the surprising connections you didn't know were there, the 'why' mined from NOTE, WHY, and HACK comments, and a set of suggested questions worth asking your graph. graph.json is the raw structure, ready for anything downstream.

Code is parsed locally with tree-sitter — real AST extraction across 36 languages, no model call required. For everything that isn't code — docs, PDFs, SQL, Postgres schemas, Terraform — Graphify uses your model to read it into the graph. And 'your model' means yours: Claude, OpenAI, Gemini, DeepSeek, Kimi, Bedrock, Azure, or Ollama if you want the whole thing running locally.

Query, don't grep

Once the graph exists, your assistant stops guessing. graphify query answers questions over the structure, graphify path traces how two pieces of code connect, and graphify explain walks you through why a node matters. An MCP server (stdio, plus a shared HTTP mode) exposes eight tools, so the graph is available natively in 17 assistants — Claude Code, Cursor, Copilot, Codex, Gemini CLI, Aider, Amp, Devin, and more.

There's PR intelligence too: graphify prs pulls CI and review status, triages open PRs with AI, and flags merge-conflict risk by checking which PRs touch the same graph communities — not just the same files.

Grounded, not vibes

Every relationship in the graph carries a confidence tag: EXTRACTED (found directly in the AST), INFERRED (connected by the model), or AMBIGUOUS (couldn't be fully resolved). When your assistant answers from the graph, the answer traces to a real path through real files — and you know exactly what was found versus what was guessed.

Everything runs on-device. Your code never leaves your machine unless you point Graphify at a hosted model, and there's no telemetry. When you want the graph elsewhere, export it to Neo4j, FalkorDB, GraphML, or Obsidian.

Get started in two minutes

Install the CLI with uv tool install graphifyy, run graphify install to wire it into your assistant, then type /graphify . in your next session. A few minutes later you'll have a map of your codebase your assistant can actually reason over — and so can you.

Graphify is MIT-licensed and built in the open. Point it at your own repo and tell us what the graph shows you.