It's a playground for teams to experiment with LLMs and prompts, track a history of what they've tried, and add evaluations. It helps developers iterate towards a working prompt and model combination for their use case more efficiently, and ships with an SDK.
