Memtrace provides a persistent, bi-temporal, and episodic knowledge graph specifically designed for coding agents, built directly from your codebase's Abstract Syntax Tree (AST). Unlike vector similarity or semantic chunking, it creates a precise structural representation where every function, class, and API endpoint is a node with deterministic, typed relationships. This enables agents to perform rapid graph traversals for tasks like caller/callee identification, blast radius analysis, and understanding architectural evolution over time. All processing, including parsing, graph construction, embedding generation, and querying, happens entirely on your local machine, ensuring privacy and eliminating API costs.
Key Features
01Bi-temporal, episodic, structural knowledge graph from AST
02Temporal memory with version history and evolution scoring
03Cross-service HTTP API topology mapping between repositories
04Rust-native compiled binary for sub-millisecond query latency
05Hybrid search combining BM25, vector embeddings, and Reciprocal Rank Fusion
06109 GitHub stars
Use Cases
01Providing coding agents with a deep, structural understanding of codebases
02Analyzing code changes over time to understand architectural evolution
03Identifying dependencies, tracing call graphs, and spotting dead code
