Verantyx

Small router always on. Larger models only when needed. Memory survives restarts.

Verantyx is a local AI operations harness. It controls which model is called, when memory is written, and how internal agreement is handled. The router stays local; larger speaker models are called only when needed.

60-second demo (no model weights required)

git clone https://github.com/Ag3497120/verantyx-cli.git
cd verantyx-cli
python3 -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt
python3 scripts/smoke_router_classify.py --no-model

This smoke test checks routing only; it does not measure answer accuracy.

Full quickstart

With router weights available, start the CLI with:

source .venv/bin/activate
python3 verantyx.py

For first-time setup and the weight conversion path, see docs/QUICKSTART.md.

Honest boundaries

Routing, memory, and council are control structures; they are not world knowledge or an accuracy booster. See benchmarks/README.md for measured results and limits.

Documentation

• Quickstart
• Japanese guide
• Omni profiles
• Demo
• Contributing