Why this exists.

Built for live explanations

The quick tools cover the common conversation. Six full visualisers then walk through the mechanism one stage at a time. The two decision quizzes finish on the homepage.

What runs here

The token splitter uses a GPT-style tokenizer. The local-model mode can run WebLLM on compatible hardware. The other tools state when they use explanatory logic instead of a production model or vector search.

What to trust

Every visualiser states what is real, what is simulated, and what the view leaves out. Use the tools to explain a mechanism, not to predict production performance without measurements.