EvoMap Open Sources AutoResearch for AI Agent Experiments

EvoMap announced in a press release that it has released AutoResearch, an open source system for AI agents to move research ideas from hypothesis to experiment and then use evidence to decide the next step.

AutoResearch uses multiple AI models to generate and cross review research ideas. Accepted ideas are turned into executable research plans with metrics, success criteria, resource budgets and evaluation steps. Separate agents handle planning, implementation, experimentation, analysis and review.

The system stores research state, code, experiment logs, metrics, failures and decisions in a persistent workspace. EvoMap said this lets AutoResearch continue unfinished work rather than restart from the beginning. Independent and blind review is used before a research direction can be closed.

EvoMap said AutoResearch was tested on a Django issue from SWE-bench Lite, where it moved from 2 of 7 to 4 of 7 on official new feature tests, then reached 7 of 7 while keeping 203 of 203 regression tests passing. On the RSICD benchmark, an AutoResearch generated idea improved mean Recall from 32.84 to 34.69.

The AutoResearch code is available on GitHub, and the paper, "AutoResearch: Insight In, Hallucination Out," is available on arXiv.