RD-Agent is a Python package for automating data-driven R&D workflows, with Docker and Conda-based setup, and LiteLLM/DeepSeek integration support. Latest releases span v0.8.0 down to v0.5.0, with features related to data science and multi-agent workflows.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
RD-Agent is a Python project intended to automate industrial R&D processes by coordinating data-driven workflows through agents. The README describes multiple focus areas, including data mining, automatic reading of research papers, and autonomous model/data development, with demonstrations and a live demo link.
How it works
The project implements a framework with agents (R for proposing ideas and D for implementing them) to automate data-driven R&D tasks. It includes a data science workflow and a data-centric quant finance variant (RD-Agent(Q)). It supports multi-agent loops (e.g., data mining, model looping) and integration with LLM backends through LiteLLM, DeepSeek, and others. It provides a Quick Start with Docker, conda environments, and a dev setup for contributors. Health check commands and configuration options are described.
Getting started
-
Docker installation recommended for demos: reference to docker run hello-world for verification.
-
Conda environment creation:
conda create -n rdagent python=3.10 -
Activate the environment:
conda activate rdagent -
Install RD-Agent (user):
pip install rdagent -
Developer installation from source:
git clone https://github.com/microsoft/RD-Agent cd RD-Agent make dev -
Health check:
rdagent health_check --no-check-env -
Configuration mentions support for various chat/embedding models and backend options:
- LiteLLM as default backend with multiple configuration approaches (Unified API base, separate chat/embedding bases).
- DeepSeek experimental support with configuration example in .env.
Recent releases
- v0.8.0 (2025-11-03) with features such as add a rag mcp in proposal.
- v0.7.0 (2025-07-08) with features like add code change summary.
- v0.6.1 (2025-06-28) and v0.6.0 (2025-06-26) include bug fixes and async mechanisms for multi-trace.
- v0.5.0 (2025-06-18) includes features like add a check for NaN values in score_df.
Traction
- Stars: 14131
- Forks: 1808
- Open issues: 195
Behind the repo
- Organization: Microsoft
- Repository: RD-Agent
- License: MIT
- Language: Python
- Created: 2024-04-03
- Last push: 2026-08-04
Caveats
- Platform note: RD-Agent currently only supports Linux.
- Requires Docker for most demos; health check checks Docker installation and port occupancy.
- Development setup is described for contributors; active CI and pre-commit hooks are present in the repo.






