colbymchenry/codegraph

CodeGraph – Semantic Code Intelligence for AI Agents

What it is – CodeGraph is a locally‑run service (built in Rust) that builds a knowledge graph of every symbol, call edge and dependency in a codebase. The graph is stored in a hidden .codegraph/ folder and kept up‑to‑date as files change.

Why it matters for LLM‑powered tools – When an AI‑assistant (Claude Code, Cursor, GitHub Copilot, etc.) needs to answer a question or make a change, it normally has to discover the code structure by grepping, reading files and stitching call‑paths together. That costs many tool calls, tokens and time. CodeGraph gives the agent a single, pre‑computed payload that contains the exact symbols and call‑chains relevant to the request, dramatically reducing discovery work.

Key capabilities

• Full‑stack language support – 30+ languages (TS/JS, Python, Go, Rust, Java, C/C++, Swift, etc.) are parsed automatically; no per‑language configuration is required.
• Cross‑file, dynamic‑dispatch resolution – the graph captures calls that static greps miss, so agents get true call‑graph context.
• Live updating – a file‑watcher keeps the graph fresh; agents always query the latest state without re‑indexing.
• Zero‑install runtime – the CLI bundles its own runtime, works on macOS, Linux and Windows, and can also be installed via npm (@colbymchenry/codegraph).
• Agent integration – a single codegraph install command wires the service into supported agents (Claude Code, Cursor, Codex, Gemini, GitHub Copilot, etc.) via an MCP (multi‑client protocol) server.
• CLI workflow – codegraph init creates the index, codegraph upgrade updates the tool, and codegraph uninstall removes all agent configurations.

Performance evidence – Benchmarks on seven real‑world repos (VS Code, Excalidraw, Django, Tokio, OkHttp, Gin, Alamofire) show:

• 44 % lower cost and 62 % fewer tokens on average.
• 88 % fewer tool calls (often down to a single codegraph_explore call).
• 0 file reads when the graph is present, compared with up to 43 reads without it.
• Speedups from 1.4× to 3.6× per query.

How to try it

1. Install the CLI (curl script or npm i -g @colbymchenry/codegraph).
2. Run codegraph install to connect your AI agents.
3. Inside a project, run codegraph init – this builds the graph and creates .codegraph/.
4. The agents will now automatically use the graph for code‑related queries.

Where to learn more – Full docs, API reference and a website are hosted at https://colbymchenry.github.io/codegraph/. A wait‑list for the hosted SaaS version is available at https://getcodegraph.com.

---

TL;DR – CodeGraph is a Rust‑based, language‑agnostic code‑graph builder that plugs into popular LLM coding assistants, giving them a single, up‑to‑date knowledge source. This cuts discovery overhead, reduces token usage and speeds up answers, making AI‑driven code assistance more efficient and cheaper.