ASI-Evolve: an open-source AI agent that runs its own research
Chinese researchers just dropped ASI-Evolve, a 100% open-source agent framework designed to conduct AI research autonomously. It reads papers, designs experiments, runs code, analyzes results, and iteratively improves itself through a closed-loop cycle. Both code and paper are out.
ASI-Evolve is an open-source agentic framework built for "AI-for-AI" research, aiming to automate the entire machine learning experimentation loop. Both the codebase and the accompanying paper are fully open source.
How it works
The system operates on a closed "learn-design-experiment-analyze" cycle. Instead of just helping you summarize a paper or autocomplete code, it reads literature to understand current methods, designs new experiments, runs the code, and analyzes the results to iteratively refine its approach.
Researchers and developers can inspect the framework via its arXiv paper (2603.29640) and open-source repository.
Why it matters
Most current research assistants remain piecemeal tools for literature search or code drafting. ASI-Evolve tries to close the loop entirely, letting the agent explore hypotheses and improve itself.
We think the real test will be how reliably it runs complex experiments without human intervention. But as an open testbed for autonomous research, it gives the community a tangible reference point to build on.
Sources (1)
- arxiv.org(opens in a new tab)Originalarxiv.org