
Biomni is an open-source biomedical AI agent from Stanford's SNAP group that carries out research tasks from natural-language instructions. It combines language-model reasoning, retrieval-based planning and code execution.
AgentFree
Step 1: Run the setup script and activate the biomni_e1 conda environment.
Step 2: Install the biomni package with pip.
Step 3: Add your API keys in a .env file or as shell environment variables.
Step 4: Create an agent in Python and point it to a data path; the data lake downloads on first run.
Step 5: Call the agent with a task written in plain language.
Step 6: Optionally launch the Gradio demo or save the conversation history as a PDF.
It executes biomedical research tasks from natural-language instructions, with examples including CRISPR screen planning, scRNA-seq annotation and ADMET prediction.
The data lake is about 11GB and downloads on first run by default. You can skip it for faster setup if your tools do not need it.
The README lists several providers configured through API keys, plus custom endpoints, including a locally served Biomni-R0.
Yes. You can add MCP servers through a config file so the agent can call external tools.
BiomniBiomni is a general-purpose biomedical AI agent from Stanford that carries out research tasks across biomedical fields. It combines LLM reasoning, retrieval-augmented planning and code execution.
Cell2SentenceCell2Sentence applies large language models to single-cell RNA data by turning each cell into a text sentence of gene names. The repository holds the code and tutorials for working with these models.
Halmos LabsHalmos Labs builds an autonomous research system. You bring a scientific question and data, and it returns findings, reports and analysis files you can trace.