RAT-retrieval-augmented-thinking

Why we included this project

RAT is a command-line chat tool that works in two stages. DeepSeek first reasons through the problem and produces a thinking trace, then the model you pick on OpenRouter uses that trace to write the final answer. That setup lets you get the depth of a reasoning model while keeping the voice, style, or cost profile of a different model. The tool also keeps conversation context, lets you switch response models mid-session, and can show or hide the reasoning process, which is useful for seeing how much visible thinking changes answer quality. It installs as a lightweight Python package you run as the `rat` command, so it's a practical way to experiment with multi-model reasoning chains without building the plumbing yourself.