Biomni: Run Biomedical Research Tasks with an AI Agent

Overview

Biomni is a general-purpose biomedical AI agent that turns natural-language research requests into tool-assisted workflows. It combines language-model reasoning, retrieval-augmented planning, and code execution to help with tasks across biomedical fields, from experimental planning to data analysis.

It is intended for biomedical researchers and developers who want an extensible research assistant, rather than a single-purpose analysis package. Its broad scope is useful when a task can draw on multiple tools or knowledge sources, but results still need expert review.

Key Features

• Accepts biomedical tasks in natural language through the A1 agent interface.
• Plans and executes tasks using language-model reasoning and code-based tools.
• Retrieves relevant resources from a biomedical data lake and a curated know-how library.
• Supports multiple model providers and custom OpenAI-compatible endpoints.
• Integrates external tools through Model Context Protocol (MCP) servers.
• Includes Biomni-Eval1, a benchmark dataset for biological reasoning tasks.
• Offers a Gradio interface and can save execution histories as PDFs.

Use Cases

• Biomedical researchers can use it to draft experimental plans, such as candidate genes for a CRISPR screen, then check the suggestions against domain knowledge.
• Computational biologists can explore workflows for tasks such as single-cell RNA-seq annotation when they want an agent to coordinate tools and analysis steps.
• Drug-discovery teams can request predictions such as compound ADMET properties as an exploratory aid, not as a substitute for validated assessment.
• Developers can add biomedical tools or MCP integrations when building research workflows that need a natural-language interface.
• Researchers evaluating biological reasoning can use Biomni-Eval1 to assess responses across its included task categories.

Project Facts

• Language: Python
• License: Apache-2.0
• Stars: 3.9k
• Forks: 723
• Topics: agent, ai, biomedicine
• Archived: No

Getting Started

The environment setup is substantial. Follow the environment setup guide, then install the package:

conda activate biomni_e1
pip install biomni --upgrade

Configure an API key for your chosen model provider before initializing an agent. See the repository README for provider options, setup details, and usage examples.

Considerations

• The default setup downloads about 11 GB of data lake files on first use. Storage, bandwidth, and environment setup are significant requirements.
• Biomni requires model-provider credentials unless configured to use a locally served or custom model.
• The README warns that the agent runs LLM-generated code with full system privileges, including access to files, network, and system commands. Use an isolated, sandboxed environment, especially with sensitive data or credentials.
• The README says the described release was frozen as of April 15, 2025, and differs from the current web platform. Check the repository for current code and setup guidance.
• Some biomedical tools, datasets, or software integrated with Biomni may have more restrictive licenses than the project itself. Review component licenses before commercial use.