DFM (Discovery Foundation Models) is neither a single model nor a software framework that forces every system through one pipeline. It is the research paradigm, capability definition and system organisation that PhAI Labs proposes around Discovery Intelligence: through formal definitions, process diagrams, system instances and real scientific cases, it turns “discovery” from a vague notion into a capability of model systems that can be learned, executed and evaluated.
DFM is not about making AI finish existing research tasks faster; it is about systems taking part in a fuller discovery process: identifying valuable unknowns, turning them into researchable problems, building and revising hypotheses, calling research tools, data and experimental environments, testing and updating judgement against external evidence, and consolidating reusable research structure and discovery capability.
This release makes the DFM technical report and the DFM research-map repository (GitHub) public. The repository carries the conceptual structure of the Discovery Loop, research directions, the open-source plan, demos and how to collaborate; its directories and direction markers reserve future scope in public and do not mean the implementations are finished or released.
Over the next three days PhAI Labs releases three independent works: on 16 September ScienceBuddy, an AI research partner and interactive Scientific Agent exploring Recursive-in-Recursive Self-Improvement (RSI in RSI) in real research interaction; on 17 September ScienceIDE, a science integrated development platform bringing scientific environments, code, tools and training pipelines together; on 18 September JEPA-Anything, a cross-domain unified science world model for state prediction, intervention simulation and scientific diagnosis. ScienceBuddy, ScienceIDE and JEPA-Anything are three fully independent modules and research directions: independent in product form, usage, papers and implementation, with no required calls or dependencies between them. Looking ahead, the three could form a composable research path: ScienceBuddy collects feedback and research trajectories from the interaction between scientists and agents; ScienceIDE organises that data and experience into reusable scientific environments for task execution, reinforcement learning and model training; JEPA-Anything explores unified state prediction and simulation across multiple scientific environments and interventions. That is a possible future combination, not an integration that exists today, and it does not change their independent standing.
The DFM Scientist Collaboration Program, opened the same day, brings real and important scientific problems, high-value research data, interactive research environments, experimental and expert feedback, and real verification conditions into the paradigm. The first round concentrates on life science and biomedicine and is open to chemistry, materials and other frontier fields.