Review of turbo-graph
AI coding agents need to understand code, not documents. Standard RAG (Retrieval Augmented Generation) retrieves text chunks that 'look like' the query. For a question like 'where is this function called from?', you don't need similar text - you need the call graph, the import graph, the type graph.
turbo-graph builds a graph of code entities (functions, classes, modules) and their relationships (calls, imports, extends, uses). Combined with vector search via TurboQuant, it gives coding agents the structured understanding they actually need.
Vector search alone is great for 'find code that looks like X' but bad for 'find what depends on X'. Graph queries are great for dependencies but bad for fuzzy semantic matches. turbo-graph gives you both: vector search to find candidate code, graph traversal to follow relationships. The result is much higher precision on real coding questions.
Works as a memory backend for Claude Code, Cursor, Aider, and other agent harnesses. The agent queries turbo-graph the same way it would query any other memory layer - through standard MCP or function calls.
Solves a real problem that most AI coding tools don't even acknowledge: vector search is the wrong primitive for code understanding. If you're building serious AI coding agents and want them to reason about code structure (not just text similarity), turbo-graph is worth studying.
|