Jakedismo/codegraph-rust

What it solves

AI coding assistants often struggle to understand the overall architecture of a codebase, relying on simple keyword searches or reading files one by one. This leads to high token consumption and a lack of deep structural understanding of how different parts of the code relate to one another.

How it works

CodeGraph transforms a codebase into a semantically searchable knowledge graph. It combines Abstract Syntax Tree (AST) parsing, Language Server Protocol (LSP) resolution, and embeddings to map out relationships like function calls, dependencies, and dataflow.

It offers three indexing tiers—fast, balanced, and full—allowing users to trade off indexing speed for graph richness. The system uses a hybrid search approach (combining vector similarity, lexical search, and graph traversal) to retrieve context. This data is then served to AI agents via four specialized tools (agentic_context, agentic_impact, agentic_architecture, and agentic_quality) that use reasoning loops (like ReAct or LATS) to synthesize answers rather than just returning raw search results.

Who it’s for

Developers using AI coding assistants (such as Claude Code or Cursor) who want to provide their agents with a deeper, structural understanding of their project's architecture and dependencies.

Highlights

• Knowledge Graph Integration: Moves beyond simple embeddings by mapping actual code relationships (e.g., who calls what).
• Tiered Indexing: Flexible indexing levels based on the need for speed versus depth of analysis.
• Agentic Tools: Provides high-level tools for impact analysis, architectural overview, and quality assessment.
• Context Overflow Protection: Automatically manages token budgets and snippets to prevent LLM context window overflow.
• Hybrid Search: Combines semantic vector search with exact keyword matching and graph traversal.