Demo
Hosted demo
The Streamlit demo is prepared for Hugging Face Spaces:
The Space runs the same app as the local command below, but starts Ollama inside
the Docker container and prepares the bundled local models:
deepseek-r1:latest and qwen3:8b.
On the free Hugging Face CPU tier, the first request can be slow while the Space wakes up and Ollama loads a model. Short questions and one Ollama-backed bot at a time give the smoothest demo experience.
Local demo
Run the CLI benchmark against the default FAQ file:
python examples/run_benchmark.py
Run the same benchmark against the adversarial FAQ set:
python examples/run_benchmark.py --faq-csv data/faq_adversarial.csv
Start the Streamlit inspection app:
streamlit run app/main.py