What problem does it solve?
Autoresearch eliminates the coordination burden of running an entire AI research project by continuously orchestrating literature review, hypothesis testing, experimentation, and synthesis into publishable outputs.
Core Features & Use Cases
- Two-loop experiment orchestration: Runs an inner loop for rapid hypothesis experiments and an outer loop for reflective synthesis and direction updates.
- Domain-skill routing: Delegates execution to the appropriate research skills for data, training, evaluation, optimization, serving, interpretability, and infrastructure.
- Continuous agent operation: Sets up a wall-clock heartbeat so the research keeps progressing across time and interruptions, while tracking state in a durable workspace.
- Research memory and reporting: Maintains
research-state.yaml, findings.md, and research-log.md, and generates progress presentations and final paper output through the library.
- Use Case: Start from a vague research goal like "Which training recipe improves benchmark X" and let the skill bootstrap hypotheses, run iterative experiments with a proxy metric, synthesize patterns, and end with a paper-ready narrative.
Quick Start
Use the autoresearch skill to initialize a project workspace and begin autonomous two-loop research by running it from your research project root.
