RD-Agent

Open source

RD-Agent is a Microsoft Research framework that uses LLM agents to automate data-driven research and development, including machine learning engineering and quant trading.

GitHub stars
15k
Last 30 days
+442(+3%)
Last commit
23 days ago
Repository age
2 years
Version
v1.0.0
Licence
MIT
Self-hosted
Yes

About RD-Agent

RD-Agent aims to automate the high-value, generic parts of research and development, which it says centre on data and models, by letting AI drive data-driven AI. The repository is a Python project from Microsoft with a live demo, documentation, a technical report and several papers.

Its scenarios include a data science agent with a Kaggle scenario, an R&D-Agent-Quant framework for quantitative finance, and an FT-Agent for LLM fine-tuning. On MLE-bench, which draws on 75 Kaggle competitions, the README says it leads as the top-performing machine learning engineering agent, with an o3 and GPT-4.1 combination scoring 30.22% on all tasks.

LiteLLM is the default backend for connecting to multiple LLM providers. A newer web frontend, started with rdagent server_ui, gives real-time interaction and trace viewing, though it excludes the data_science scenario for now. The project is released under the MIT licence and discussed on Discord.

Key features

  • Kaggle data science agent scenario
  • Multi-agent framework for quantitative finance
  • LiteLLM as default backend for multiple LLM providers
  • Web frontend for live interaction and trace viewing
  • LLM fine-tuning agent with its own guide
  • Benchmarked on MLE-bench with detailed run traces

Good fit for

  • Kaggle-style data science tasks
  • Quantitative finance research
  • Automating LLM fine-tuning experiments
Built with
Python tools
Tags
llm
agents
data-science
automation
machine-learning
kaggle
quant
research

RD-Agent: questions and answers

What is RD-Agent used for?
RD-Agent is a Microsoft Research framework that uses LLM agents to automate data-driven research and development, including machine learning engineering and quant trading. It is a good fit for kaggle-style data science tasks, quantitative finance research and automating LLM fine-tuning experiments.
Is RD-Agent open source?
Yes. RD-Agent is open source under the MIT licence. Its source code is on GitHub at microsoft/RD-Agent and is written mainly in Python.
Is RD-Agent free?
Yes. RD-Agent is open source, so the software itself is free to use.
What are some alternatives to RD-Agent?
Similar open-source tools in the AI Infrastructure category include TPOT, LocalAI and InternVL. SaaS products in the same category include TinyTune, Vellum and Amazon Bedrock.
Is RD-Agent actively maintained?
Yes. The most recent commit to RD-Agent was on 21 October 2026, and the latest release is v1.0.0, published on 23 September 2026. The project has 15k stars on GitHub, with 442 added in the last 30 days.

Open-source alternatives to RD-Agent

See all

SaaS alternatives to RD-Agent

See all