OpenScientist
OpenScientist provides an automated AI scientist platform called Praxis R1 that takes a research question in plain language and drives it through the full scientific method—from literature review and hypothesis generation to experiment execution and paper writing. The system orchestrates swarms of specialist agents that run on cloud GPUs, with users able to choose between fully autonomous or stage-by-stage guided runs.
- Artificial Intelligence
- AI Agents
- Drug Discovery & Therapeutics
- Biotechnology
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Scientific discovery is bottlenecked by human time and manual workflows, limiting how many hypotheses can be tested and slowing the pace of research. Researchers often spend more time on literature review, tool orchestration, and report writing than on actual scientific reasoning.
Solution
Praxis R1 is an automated AI scientist that independently drives research from hypothesis generation through execution to written briefs. The system plans research objectives, searches literature databases, connects to structural docking servers, and evaluates candidate results using multi-objective criteria. It improves over time through reinforcement learning, rewarding valid hypotheses and successful tool selections while penalizing failed paths. Each completed research thread sharpens the model's policy for subsequent experiments, creating a compounding effect where results inform future decisions. The platform supports both fully autonomous and guided runs, with a stage-by-stage approval workflow. Outputs include self-authored research briefs with confidence scoring and verified citations, requiring zero human drafting.
Target Audience
Primary users are research scientists, biotech and pharmaceutical R&D teams, and academic laboratories seeking to scale hypothesis testing and accelerate drug discovery through automated experimentation. The platform also serves curious non-experts who can ask questions in plain language.
Features
- Autonomous hypothesis generation with parallel research thread management across any field, including biology, physics, machine learning, and economics
- Dynamic agent swarm orchestration
- a recorded run split into 30 specialist agents across 5 teams working in parallel
- Multi-objective candidate evaluation scoring binding, stability, and developability for molecular research
- Reinforcement learning environment that rewards valid scientific decisions and penalizes failed pathways
- Self-authored research briefs with confidence scoring, verified citations, and LaTeX-compiled PDF output
- Private cloud sandbox per investigation with cloud GPUs metered by the second
- Seven-stage pipeline from clarification and literature mapping to blueprint, execution, and final paper