🛠️ Benzi: Compiler-Backed AI Coding Agent

Benzi takes a different approach compared to most AI coding assistants on the market. Instead of simply feeding code snippets to a large language model, it first acts like a compiler, resolving the entire codebase into a queryable map of call graphs, data flows, and control flows. The hard part of code analysis is that static methods often miss dynamic behaviors, so Benzi includes a runtime tracer to settle what static analysis cannot. It also functions as an MCP server. This approach of combining traditional compiler principles with LLMs is a highly interesting exploration into improving the accuracy of AI-driven code edits.