Tool builds code knowledge graphs

A new open-source tool called graphify turns a codebase into a queryable map, letting coding assistants inspect structure instead of rereading raw files. (github.com) The project is maintained by Safi Shamsi and, as of April 14, 2026, its GitHub repository showed about 23,500 stars, 2,500 forks, and a latest release, version 0.4.1, published within the past two days. (github.com, github.com) Graphify says it delivers 71.5 times fewer tokens per query than feeding raw files to a model, and it works through slash commands in tools including Claude Code, OpenAI Codex, OpenCode, Cursor, Gemini Command Line Interface, GitHub Copilot Command Line Interface, Aider, OpenClaw, Factory Droid, and Trae. (github.com, graphify.net) A code knowledge graph is a database of relationships: which files call which functions, which modules depend on which classes, and which documents explain which systems. Graphify builds that map so a model can ask for a path, a dependency chain, or an explanation without loading whole directories into context. (graphify.net) The software combines two kinds of reading. Tree-sitter, a parser used to break source code into syntax trees, extracts structure from languages including Python, TypeScript, Go, Rust, Java, C, C++, Ruby, C sharp, Kotlin, Scala, PHP, Swift, Lua, Zig, PowerShell, Elixir, Objective-C, and Julia, while language and vision models pull concepts from prose, images, diagrams, and screenshots. (github.com, graphify.net) Those pieces are merged into a NetworkX graph, grouped with the Leiden clustering algorithm, and exported as three main outputs: an interactive `graph.html`, a machine-readable `graph.json`, and a text report called `GRAPH_REPORT.md`. (graphify.net) The pitch lands at a moment when coding agents are running into context-window limits on large repositories. Instead of sending thousands of lines back into a model on every turn, graphify stores a smaller structural summary that can persist across sessions. (github.com, graphify.net) The tool is broader than source code alone. Its current release notes say version 0.4.0 added local transcription for video and audio files through Faster Whisper, with transcripts cached so reruns skip files already processed, and the README says users can also drop in Portable Document Format files, markdown, whiteboard photos, and images in other languages. (github.com, github.com) Recent fixes also show the trade-offs in automating code understanding at scale. Version 0.4.1 corrected an ignore-file bug that could make a project scan 40,000 files instead of about 200, and another bug that could cut extracted graph nodes by as much as 2.4 times when a read-only subagent was used. (github.com) Graphify is not the only attempt to give coding assistants memory, but its approach is unusually explicit: build a map first, then query the map. For teams struggling with large repositories and rising token bills, that shifts the bottleneck from rereading files to maintaining a clean graph of what the code actually is. (github.com, graphify.net)