oasis

Why we included this project

OASIS is a security auditing tool that runs Ollama models against your code to find potential vulnerabilities, and it keeps the whole analysis on your own hardware. The scan pipeline is built on LangGraph, so you can point lightweight models at the first pass and save the heavier ones for the parts that need a deeper look. What stands out is the finding-validation step: rather than taking the model's word for it, OASIS runs a code-driven check and returns an exploitability verdict backed by citations, with the model's narrative forced to stay consistent with that result. Reports come out as canonical JSON with HTML, PDF, and Markdown versions, and a password-protected web dashboard lets you filter by severity or project and ask an assistant questions that draw on a local embedding cache. If you want AI-assisted code review without shipping source code to a third-party API, this local-first setup is a practical fit.