2026-10-10 · America/Los_Angeles · 社区动态 · #12

I use a ~322M local model to judge every tool call my coding agents make (about 90 ms on CPU, no API)

Coding agents run shell commands constantly. I wanted each one checked before it executes, and I didn't want to send every command to a paid API. laya-guard runs on your machine and answers allow, review, or block for each tool call. Regex policies catch known-dangerous patterns first (rm -rf /, curl … | sh, secrets, auth bypass). Plain commands like ls, git status and npm test skip the model, but only when there's no piping, chaining or redirecting, no -exec -style flag, and no sensitive path. Everything else is judged by Laya (~322M, Apache-2.0) on the CPU in about 85 to 90 ms. The results…

热度 44.8 / 100；排名与评分保留该期记录。