Comments2

  • @2501pr0ject
    Abdelatif Touati
    Local-first RAG, LLM evaluation & agents · Maintainer of RAGnarok-AI · Building reliable, privacy-first AI systems
    2m ago

    A security harness for tool-calling agents. It acts as a deterministic gateway between an agent and its tools, designed to prevent any action classified as irreversible from executing without a matching signed approval — even if the agent runtime is fully compromised, provided the isolation boundary and trusted components remain intact.

    github.com/2501Pr0ject/Ullr-AI

  • @2501pr0ject
    Abdelatif Touati
    Local-first RAG, LLM evaluation & agents · Maintainer of RAGnarok-AI · Building reliable, privacy-first AI systems
    just now

    I'm also training my own LLM-asJudge but it's still private. I think next week i'll turn it inot a public repo.

Log in to join the thread.