Wow model sekecil ini bisa dapet score 89%, mantep juga ini model, speed-nya ngeri 355 tok/s! 🤯
Tool calling oke, cuman sayang kemampuan codingnya sama memory recall-nya lemah.
Model seperti ini cocoknya buat entity extraction, anomali detection, & kerjaan2 kecil cepat lainnya.
AI coding agents still waste a lot of time repeatedly scanning files, running ad hoc grep searches, and losing context in larger codebases.
I open-sourced Lexa to help solve that
https://t.co/qiR7xCPFI8
Lexa is a local code intelligence engine for AI coding agents. It indexes your codebase once, then exposes MCP tools that Claude Code, Codex, and OpenCode can use to search symbols/text, inspect files, trace dependencies, read focused context, and edit files more safely.
The goal is simple, give coding agents a more stable way to understand your local codebase.