PhAI Labs

Advancing Discovery Intelligence through models, scientific data, and learning environments.

PhAI Labs is a frontier AI research lab working on foundation models, reinforcement learning, and scientific agents. We build systems that learn from real tasks, experimental evidence, and researcher feedback—and carry that experience into future discoveries.

Our work includes:

• Discovery Foundation Models — Our research framework for making open-ended discovery a learnable, executable, and evaluable capability.
• ScienceIDE — Executable scientific environments for agent learning and evaluation, with the PhAI-IDE-4B / 9B / 72B model series.
• ScienceBuddy — An interactive scientific workspace exploring recursive-in-recursive improvement of agent harnesses and models.
• JEPA-Anything — Research into predictive world models across scientific domains.

We collaborate with scientists, research teams, and experimental platforms to connect model learning with meaningful scientific problems and real-world validation.

Every exploration should become a starting point for the next discovery.

🌐 Website · 📄 Research · 🤝 Scientist Collaboration · ✉️ Contact