Describe a research problem at 11pm. Wake up to a publication-ready paper — literature surveyed, experiments run, approaches evolved across generations.
$ pip install git+https://github.com/mohitkr/autonomous-researcher.git
Each agent specializes in one phase. Between generations, the best approaches survive and mutate. The worst are discarded.
The full research lifecycle, from literature review to publication draft.
Multiple approaches run in parallel, best performers survive and mutate, weaker ones are eliminated. Like a genetic algorithm for research ideas.
Parallel experiment execution on AWS. Manages training jobs, hyperparameter tuning, and result collection automatically.
Sleep prevention, crash recovery, and resume from any phase. Start before bed, wake up to results.
Real-time web UI showing generation progress, experiment metrics, and agent activity as it happens.
Each agent remembers what worked and what failed across generations. Better decisions compound over iterations.
Consistent train/val/test splits and metrics. Every candidate is compared on equal footing.
A dedicated reviewer agent critiques results each generation, catching methodological issues before they compound.
Starts by surveying relevant papers, extracting techniques and baselines. Approaches build on the state of the art.
Claude, Gemini, or DeepSeek. Mix fast models for search with powerful models for ideation. Configure per agent role.
Use one provider or mix models across agent roles for the best cost-quality tradeoff.
Install, configure your LLM provider, describe your problem.
~/.autonomous-researcher/config.yaml.A complete research paper with methodology, experiments, and results.