UCSF BioRouter is an AI-powered integrated research environment that unifies commercial, institution-hosted, and local LLMs, AI agents, Information Commons databases, and customizable workflows into one extensible tool for explorative analysis, prototyping, automation, and federated cross-institution collaboration.
Think of BioRouter as your intelligent research co-pilot — one that can read papers, query databases, run code, manage files, and carry out complex multi-step research tasks, all from a single unified interface.
- Multi-Provider LLM Support — Works with Anthropic Claude, OpenAI GPT, Google Gemini, Amazon Bedrock, Azure OpenAI, Ollama (local), and many more
- UCSF Institution-Ready — Built-in support for UCSF-hosted ChatGPT (Azure) and UCSF-hosted Anthropic (Bedrock)
- Extensible via MCP Agents — Connect to databases, web tools, file systems, APIs, and custom agents through the Model Context Protocol
- Recipes & Automation — Package any workflow into a shareable, reusable recipe; run it on a schedule without lifting a finger
- Fully Local Mode — Use Ollama for completely air-gapped, private inference — nothing ever leaves your machine
- Skills System — Teach BioRouter your team's workflows and best practices as reusable instruction sets
- Desktop & CLI — Beautiful GUI desktop app for interactive work; CLI for scripting and automation
Native installers for all major platforms are available in every release:
| Platform | Package |
|---|---|
| macOS (Apple Silicon) | .zip — unzip and drag to /Applications |
| Windows (x64) | .zip — unzip and run BioRouter.exe |
| Linux Ubuntu / Pop!_OS (x64) | .deb — sudo dpkg -i biorouter_*.deb |
| Linux Fedora / RHEL (x64) | .rpm — sudo rpm -i BioRouter-*.rpm |
Always install the newest version for the latest features and fixes.
1. Download and install BioRouter from the Releases page.
2. Connect a provider — On first launch, BioRouter will walk you through choosing an LLM provider:
- UCSF users: Select Azure OpenAI (UCSF ChatGPT) or Amazon Bedrock (UCSF Anthropic)
- With your own API key: Enter your Anthropic, OpenAI, or Google API key directly
- Fully local: Install Ollama and select it — no API key needed, no data leaves your device
3. Start exploring — Type a question, describe a task, or load a recipe. BioRouter takes it from there.
- Researchers who want to analyze data, review literature, and run bioinformatics pipelines with the help of AI
- Clinicians and data scientists who need secure, institution-compliant AI access for sensitive research tasks
- Developers and engineers building automated workflows, pipelines, and agentic systems
- Teams looking to share reusable AI workflows as recipes across their organization
BioRouter routes your inputs to an LLM provider. For patient data, PHI, or other sensitive research data:
- Use institution-managed services (UCSF Azure OpenAI or UCSF Amazon Bedrock) or fully local Ollama
- Do not use personal commercial API keys with patient data
- Always verify with your institution's compliance office before processing sensitive data
See the Data Privacy Guide for full details.
In-depth documentation is available in the documentation/ folder:
| Guide | Description |
|---|---|
| Architecture | How BioRouter is built — backend, frontend, agent loop |
| Providers & Models | All supported LLM providers and models |
| Extensions, Skills & MCP | Adding tools, agents, and reusable skills |
| Recipes | Creating and sharing automated workflows |
| Schedulers | Running recipes on a schedule |
| Installation & Setup | Step-by-step setup guide |
| Data Privacy | Guidelines for handling patient and sensitive data |
BioRouter's agentic coding environment was developed with reference to the following open-source AI coding tools — we are grateful to their authors and communities:
- Goose — CLI/Desktop agent for full developer workflows (Block) — BioRouter's primary upstream foundation
- Aider — Open-source, Git-native CLI AI coding agent
- Cline — Open-source interactive CLI coding agent
- OpenCode — Open-source coding agent with multi-session and multi-provider support
- ForgeCode — Terminal AI coding assistant for task planning and code generation
If you use BioRouter in your research, please cite:
@software{biorouter2025,
title = {UCSF BioRouter: An AI-Powered Integrated Research Environment},
author = {Gu, Wanjun, Bellucci, Gianmarco and Baranzini, Sergio E.},
year = {2025},
url = {https://github.com/BaranziniLab/BioRouter}
}UCSF BioRouter is developed by Wanjun Gu (wanjun.gu@ucsf.edu) at the Baranzini Lab, Department of Neurology, UCSF Bakar Computational Health Sciences Institute. Development is supported by UCSF IT and Information Commons.
Licensed under the Apache License 2.0.