RD-Agent
An AI agent framework that automates parts of data science and R&D work, such as ML competitions and quant res
- Stars
- 15.2k
- Forks
- 2k
- Licence
- MIT
- Language
- Python
- Last update
- 16 days ago
- Type
- AI model / agent
What it is
R&D-Agent is a Microsoft open-source framework that uses large language model agents to automate research and development tasks, with a focus on data and models. The README describes it as letting AI drive data-driven AI work. The README also shows it applied to machine learning engineering benchmarks (MLE-bench) and to quantitative trading strategy research (RD-Agent(Q)).
- Run R&D-Agent on MLE-bench machine learning engineering tasks, per the linked guide
- Explore the web UI for real-time interaction and trace viewing (the README notes it excludes the data_science scenario)
- Run the Kaggle or general data science agent scenarios described in the news section
- Apply the RD-Agent(Q) framework to quantitative trading factor research
Great for
- Researchers who want to test automated ML engineering agents on benchmark-style tasks
- Data scientists exploring agent-driven model and feature development
- Quant researchers interested in automated factor and strategy research
- Teams comparing LLM backends for autonomous experimentation
Not ideal for
- People wanting a simple, ready-made app with no setup, since the README describes Python packages, documentation, and configuration rather than a finished end-user product
- Users who need results without paying for LLM API usage, since the agents rely on language model backends
- Non-Linux users, since the PyPI badge lists the platform as Linux
How to get it
- 1
Install with a package manager
You'll need: Python (version range listed on PyPI; not confirmed in the truncated README) · An LLM API key or backend (implied by LiteLLM and LLM agent usage)
Install from PyPI (package named in badge)
pip install rdagent - 2
Run it
- Install the rdagent package from PyPI
- Configure an LLM backend according to the documentation
- Open the live demo or documentation to see example scenarios
- Follow the MLE-bench or Kaggle scenario guide to run a first task
- 3
Get the most out of it
- Read the documentation at rdagent.readthedocs.io for scenario-specific setup
- Use the web UI (rdagent server_ui) to view traces, as noted in the news section
- Review the detailed MLE-bench runs published online to understand agent behavior
- Try the RD-Agent(Q) quant scenario if you work in financial research
Downloads
Strengths
- Backed by Microsoft Research with linked technical report and peer-reviewed papers cited in the README
- Documented benchmark results on MLE-bench, with run details published online
- Supports multiple LLM providers through LiteLLM, per the README
Trade-offs
- Results and costs depend on the LLM backends used, and the README reports benchmark numbers with standard deviations rather than guarantees
- Installation and configuration details were not fully visible in the truncated README, so setup may take effort
- Some scenarios are limited; the web UI currently excludes the data_science scenario
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| RD-Agent (this) | 15.2k | AI model / agent | advanced | 16 days ago | Researchers who want to test automated ML engineering agents on benchmark-style tasks |
| deep-research | 19.9k | AI model / agent | intermediate | 7 months ago | Developers who want a readable example of a deep research agent to learn from or extend |
| UFO | 10.2k | AI model / agent | advanced | 6 days ago | Developers experimenting with LLM-driven Windows desktop automation |
| huginn | 50.3k | App | intermediate | 4 days ago | People who want IFTTT-style automations without sending their data to a third-party service |
| ML-From-Scratch | 33.3k | Learning resource | beginner | 3.1 years ago | Students who want to see how algorithms like regression or neural networks work internally |
| awesome-datascience | 30.3k | Curated list | beginner | 5 days ago | Beginners who want a guided set of links to start learning data science |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.