Benzi redefines AI-native code intelligence by moving beyond simple context window dumping. Instead, it utilizes a real compiler, built on tree-sitter, to parse every project file into a detailed, queryable map. This infrastructure resolves symbols, call edges, references, and inheritance chains in a single pass, enabling O(1) query responses. By providing a deep structural understanding of the codebase before any questions are answered, Benzi facilitates more accurate and efficient AI-driven code analysis, generation, and modification, significantly reducing tokens read and overall cost compared to traditional AI coding agents.
主な機能
01129 GitHub stars
02Compiler-backed code intelligence with tree-sitter for multiple languages.
03O(1) query responses on a precise, resolved code map.
04Gated code edits with syntax and semantic verification and auto-revert.
05Runtime tracer for observing execution and updating the static map.
06Incremental reindexing and persistent memory for efficient updates.
07https://varianttech.net/
ユースケース
01Automated bug fixing and issue resolution in codebases (e.g., SWE-bench Verified).
02Deep code understanding and navigation for complex projects.
03Greenfielding applications from scratch using AI agents.