Benzi – Code Intelligence MCP/Harness beating Claude Code and CodeGraph (varianttech.net)

Benzi, a new code intelligence tool, has been unveiled, promising a significant advancement in how AI coding agents handle codebases. Unlike traditional approaches that merely dump repositories into context windows, Benzi compiles an entire codebase into a resolved, queryable map before any queries are made. Utilizing the tree-sitter parser, Benzi effectively resolves calls, data flows, and references across ten programming languages, improving code navigation and accuracy. This innovative architecture allows for quick, O(1) responses to queries and enables developers to trace bad values to their origins efficiently. The implications of Benzi for the AI/ML community are substantial. By integrating a robust indexing system and leveraging specific language grammars, it enhances the efficiency of code analysis and debugging. In rigorous testing with 500 GitHub issues, Benzi demonstrated superior performance compared to its competitors, requiring fewer lines of code to be read per task and operating at a lower cost. This level of performance underscores Benzi’s potential to revolutionize code intelligence and debugging, making it a viable tool for software engineers looking to streamline their workflows and improve code quality overall.