Skip to content
Get packs

Guide · Open source

How to use RD-Agent: install with pip and run it

Updated · written with AI from the project's own documentation and checked against our catalogue data

RD-Agent is an open-source framework from Microsoft that uses large language model agents to automate research and development work, with a focus on data and models. This guide covers what it is, who it suits, how to install the Python package from PyPI and where to go next. By the end you will know whether it fits your work and how to get the package running.

RD-Agentmicrosoft/RD-Agent · ★ 15.2k · MIT

Quick start

  1. 1Check your platform. Confirm you are on Linux, the platform listed on the PyPI badge.
  2. 2Check your Python version. Look up the supported Python range on the PyPI page for rdagent.
  3. 3Get an LLM API key. Obtain a key or backend for a provider you plan to use through LiteLLM.
  4. 4Install the package. Run pip install rdagent in your Python environment.
  5. 5Read the documentation. Open the official docs at rdagent.readthedocs.io to set up configuration and choose a scenario.
  6. 6Pick a scenario. Start with MLE-bench via the MLE-bench Guide, or with RD-Agent(Q) for quant research.
  7. 7Review the results. Compare agent outputs on your own tasks and check the linked detailed runs.

What RD-Agent does

RD-Agent is described in its README as a way to automate high-value generic R&D processes, letting AI drive data-driven AI work. The project is built around large language model agents.

The README highlights two applications: machine learning engineering benchmarks (MLE-bench) and quantitative trading strategy research through RD-Agent(Q), which it calls a data-centric, multi-agent framework for factor-model co-optimization.

  • Machine learning engineering experiments on MLE-bench tasks
  • Automated factor and strategy research for quantitative finance (RD-Agent(Q))
  • Agent-driven model and feature development

Check the requirements before you install

RD-Agent is a set of Python packages, documentation and configuration, not a finished end-user app. The PyPI badge lists the platform as Linux, so Linux is the platform the README points to.

You need a Python environment. The supported version range is listed on PyPI; the truncated README does not confirm it, so check the PyPI page before installing.

The agents rely on language model backends, so you need an LLM API key or backend. The README mentions LiteLLM as the default backend for integrating multiple LLM providers. Expect to pay for LLM API usage.

  • Linux (per the PyPI platform badge)
  • Python, in the version range shown on PyPI
  • An LLM API key or backend, used through LiteLLM

Install RD-Agent from PyPI

The README gives the package name as rdagent on PyPI. Install it with pip in the Python environment you plan to use.

After installing, the README's documentation link (rdagent.readthedocs.io) describes the scenarios and configuration in more detail.

pip install rdagent

Run your first scenario

The README points to the documentation for setup and scenario details rather than giving a full command walkthrough. For MLE-bench runs, it links to the MLE-bench Guide, which covers running ML engineering via MLE-bench.

The README also mentions a web UI, started with rdagent server_ui, for real-time interaction and trace viewing. It currently excludes the data_science scenario. Check the documentation for the current setup steps and your LLM backend configuration before running experiments.

  • Read the MLE-bench Guide in the documentation for benchmark runs
  • Use the web UI for trace viewing, noting it excludes the data_science scenario
  • Configure your LLM backend before starting agent runs
rdagent server_ui

Understand the results you get

The README reports MLE-bench results for R&D-Agent variants alongside AIDE. It notes the complexity levels (Low==Lite, Medium, High) and that results are averages over several seeds with standard deviations. Read these as the project's reported numbers, and check the linked detailed runs yourself.

For RD-Agent(Q), the README reports results on real stock markets at a cost under $10, but treat these as the authors' claims and verify them against your own data.

  • Detailed MLE-bench runs are linked from the README
  • Benchmark numbers come from the project and depend on the model combination used
  • Compare results using your own tasks before relying on them

What people use it for

  • Testing automated ML engineering agents on MLE-bench style tasks
  • Exploring agent-driven model and feature development as a data scientist
  • Researching automated factor and strategy generation for quantitative trading
  • Comparing LLM backends for autonomous experimentation

Frequently asked questions

Is RD-Agent free to use?

The source gives the licence as MIT, which allows use of the code. The agents rely on language model backends, so you should expect to pay for the LLM API usage that runs them.

What are the requirements for RD-Agent?

You need Python in the version range listed on PyPI, an LLM API key or backend, and a Linux platform according to the PyPI badge. The README is a set of packages and configuration, not a ready-made app.

How do I install RD-Agent?

Install the package named rdagent from PyPI with pip install rdagent. The documentation at rdagent.readthedocs.io covers configuration after installation.

Does RD-Agent work on Windows or macOS?

The PyPI badge lists the platform as Linux. The README does not describe Windows or macOS support, so Linux is the safer choice.

What are alternatives to RD-Agent?

The README lists AIDE, MLE-bench, AutoGluon and Auto-sklearn as comparable projects. MLE-bench is a benchmark that RD-Agent is evaluated on, not an agent alternative in the same sense.

Is RD-Agent a finished app I can use without coding?

No. The README describes Python packages, documentation and configuration rather than a finished end-user product. A live demo and web UI are mentioned, but the web UI excludes the data_science scenario.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

RD-Agent suits researchers, data scientists and quant researchers who are comfortable with Python, Linux and paying for LLM API calls, and who want to test agent-driven experimentation. If you want a ready-made app with no setup, or you work on Windows or macOS, pick a different tool such as AutoGluon or Auto-sklearn.

Install options & downloads