CallSight - Review Agentic AI Code Generations

I built CallSight because reviewing agentic/AI-generated code often led to duplicated functions, scattered changes across 15+ files, and lost visibility into execution flows. The 'lost in the middle' & attention dilution issues crept in with larger codebases even with larger context windows. Git diff would lead me to changes across 15+ files and gradually I would lose track of my own codebase execution patterns and flows in my mind-map. The inherent problem with AI is its probabilistic nature. CallSight provides a deterministic, local lens (no cloud, no LLM) on exactly how function calls and execution patterns change. Built on Tree-sitter for accurate parsing of TS, JS, TSX, JSX, Python, and Go. Two things drove the design: PR reviews - before merging AI-generated code you want to know what breaks if a changed function fails. CallSight has an impact radius overlay - hover any function and it shows exactly which other functions are affected across files, hop by hop. Makes it much faster