Kusanagi AI

For researchers in physics & material science

Agentic search, RAG over your own PDFs, and model inspection tools. Run them in this browser with your own API key, or install the desktop suite and run everything offline.

Connect a provider Run it offline instead

Connect a provider

Keys are held in this tab only and sent straight to the provider you pick — there is no Kusanagi server in the path. Tick Remember only on a machine you control; it writes the key to disk, where anything that can read this browser profile can read it.

    Default provider for the tools:

    Launch a tool

    Each card shows whether the provider it needs is configured.

    Or run the whole suite offline

    The desktop apps run against a local Ollama model. No API key, no network, nothing leaves the machine — which is the configuration to use for unpublished data.

    Install

    git clone https://github.com/prathameshnium/Kusanagi-AI.git
cd Kusanagi-AI
pip install -r requirements.txt
python local_apps/launcher.py

    What you get

    • Orochimaru Local — RAG over your PDFs with local embeddings.
    • OneTail Desktop — multi-model chat against Ollama.
    • Visualize AI — token probability inspector.
    • Hopfield demo — associative memory playground.

    Running the web apps offline

    Serve them rather than opening the files directly — the pages carry a strict Content-Security-Policy, and file:// is not an origin it can grant anything to. Run python scripts/serve.py from the repo root and the whole suite works with no network at http://localhost:8000.