snap-stanford/Biomni ? reverse-engineered prompt

Build me a biomedical AI agent that can take natural language requests and help with research tasks like planning experiments, annotating single cell data, and estimating drug properties.

I want it to feel like a real assistant for scientists, so it should be able to reason through a task, use the right tools or data when needed, and give back useful results and hypotheses instead of just chat. Please include a simple local web interface too, so I can type a request in a browser and see the agent work.

Make it easy to configure with environment variables or a simple config file, and keep the setup and startup steps clear. If it helps, add a way to save the conversation or run as a reusable Python package. Look up current docs online if you need to, and keep the first version focused on the core biomedical agent experience.