When GitHub Copilot dropped, it was magical. Within seconds, it understood what you were trying to do and suggested the next line of code. But the magic wore off fast when you realized it had no idea about your codebase's actual architecture, naming conventions, or the context that makes your code unique.
The problem isn't AI's fault. It's a context problem.
Every time a model needs to understand your code, it either (a) re-parses files on demand, losing context across the project, or (b) tries to squeeze an entire codebase into a prompt window, hitting token limits within seconds. Neither scales.
What if we gave AI—and developers—a proper database of code relationships?
The Insight
Good code analysis has always been hard because code is relational. A function calls another function. A class inherits from a parent. An import creates a dependency. Traditional tools treat files as isolated units. Modern AI treats your codebase as raw text. Neither captures the actual graph of relationships that makes code meaningful.
We needed a system that:
- Understands code structure without forcing you to use a specific parser
- Stays fast (sub-100ms queries on massive codebases)
- Persists (index once, query forever)
- Integrates with AI (give models structured context instead of text soup)
So we built CodeGraph: an open-source graph database purpose-built for code intelligence.
How It Works
The architecture is deceptively simple:
Layer 1: Parser Agnostic Core
CodeGraph doesn't have opinions about how you parse code. You bring any parser—tree-sitter, a custom AST tool, whatever—and we handle the relationship analysis and graph storage.
Layer 2: Language Plugins
We provide battle-tested modules for Python, with Rust, JavaScript, and Go coming next. Each plugin handles the language-specific AST traversal so you don't have to. Write a parser, get graph intelligence.
Layer 3: Your Application
Call parse_file(). That's it. Your code gets indexed into a persistent RocksDB graph. Now query it in milliseconds.
let codebase = CodeGraph::new(db_path)?;
let classes = codebase.query("SELECT * FROM entities WHERE kind = 'class'")?;
let callers = codebase.query("SELECT * FROM entities WHERE calls = 'parse_file'")?;
The graph stores everything: classes, functions, imports, inheritance chains, call paths. You query it like a database because it is one. Structured. Queryable. Fast.
Why This Matters Now
We're at an inflection point. AI coding tools are getting smarter, but they're still working in the dark about your code. CodeGraph changes that.
For AI Tool Developers:
Instead of dumping raw code into prompts, expose CodeGraph as an MCP server. Now your agent can ask intelligent questions: "What calls this function?" "Where's the config parsing logic?" "What's the dependency tree?" Structured queries beat raw text every time.
For Developers:
Tired of grepping to understand your own codebase? CodeGraph makes that instant. Answer questions about architecture, find dead code, trace dependency chains—all in milliseconds.
For the Open Source Community:
We're not gatekeeping code intelligence behind proprietary parsers or closed APIs. Apache 2.0. Bring your parser. Build your tools. We handle the graph part.
What's Coming
The VS Code extension is coming soon. LSP-powered, with smart context selection so AI models only see what matters. We're expanding language support to hit the top 10 languages. And we're building MCP server integration so tools like Claude Code and Cline can natively query CodeGraph.
The long game is clear: code should be a queryable database, not a text dump. That changes what's possible.
Join Us
We're just getting started. The codebase is on GitHub (link below), we're hitting Product Hunt soon, and we're looking for contributors, feedback, and the inevitable "wait, why didn't this exist sooner?" moments.
If you've ever wished your tools understood your code better—whether you're building AI agents, writing a better IDE, or just trying to navigate a legacy system—CodeGraph is for you.
Check us out. Break it. Tell us what you'd build with code intelligence that actually works.
Links:
What would you use a code graph database for? Drop a comment—genuinely curious what problems you're trying to solve. 🚀
Top comments (0)