Bohrium + SciMaster: Building the Infrastructure and Ecosystem for Agentic Science at Scale

TL;DR

The paper addresses the challenge of scaling agentic science by proposing Bohrium+SciMaster, an infrastructure-and-ecosystem stack that makes Reading, Computing, and Experiment executable as traceable services. It introduces a scientific intelligence substrate comprising a hierarchy of foundation components, a shared knowledge base (SciencePedia), and open communities (DeepModeling) that can be composed and executed on the stack. Through eleven mature master agents spanning domains, the work demonstrates substantial reductions in end-to-end cycle times and production-scale, execution-grounded signals, illustrating how a federated, governed ecosystem can enable Science-as-a-Service. Overall, the Bohrium+SciMaster framework aims to create a platform where tools, data, models, and human expertise co-evolve, enabling auditable, reusable, and scalable agentic science at community scale.

Abstract

AI agents are emerging as a practical way to run multi-step scientific workflows that interleave reasoning with tool use and verification, pointing to a shift from isolated AI-assisted steps toward \emph{agentic science at scale}. This shift is increasingly feasible, as scientific tools and models can be invoked through stable interfaces and verified with recorded execution traces, and increasingly necessary, as AI accelerates scientific output and stresses the peer-review and publication pipeline, raising the bar for traceability and credible evaluation. However, scaling agentic science remains difficult: workflows are hard to observe and reproduce; many tools and laboratory systems are not agent-ready; execution is hard to trace and govern; and prototype AI Scientist systems are often bespoke, limiting reuse and systematic improvement from real workflow signals. We argue that scaling agentic science requires an infrastructure-and-ecosystem approach, instantiated in Bohrium+SciMaster. Bohrium acts as a managed, traceable hub for AI4S assets -- akin to a HuggingFace of AI for Science -- that turns diverse scientific data, software, compute, and laboratory systems into agent-ready capabilities. SciMaster orchestrates these capabilities into long-horizon scientific workflows, on which scientific agents can be composed and executed. Between infrastructure and orchestration, a \emph{scientific intelligence substrate} organizes reusable models, knowledge, and components into executable building blocks for workflow reasoning and action, enabling composition, auditability, and improvement through use. We demonstrate this stack with eleven representative master agents in real workflows, achieving orders-of-magnitude reductions in end-to-end scientific cycle time and generating execution-grounded signals from real workloads at multi-million scale.

Figures (4)

• Figure 1: Infrastructure and ecosystem around Bohrium+SciMaster. Bohrium turns bare scientific assets (data, software, compute, and laboratory equipment) into agent-ready capabilities for Reading, Computing, and Experiment with unified interfaces, observability, and governance. It also hosts a capability and tool registry for standardized packaging and governed execution of large collections of reusable tools and workflows. Above this execution substrate, a scientific intelligence substrate provides reusable building blocks for scientific reasoning and action, including a hierarchy of scientific models (e.g., Innovator, domain foundation models, and pipeline/application models) and a knowledge substrate (SciencePedia). Open AI4S communities---represented here by DeepModeling---contribute across the stack and provide reusable open-source assets that can be integrated into agent-ready workflows on top of shared infrastructure. SciMaster orchestrates these capabilities into long-horizon, tool-augmented, multi-agent workflows, while execution traces and distributed validation signals support continual refinement at ecosystem scale.
• Figure 2: Bohrium as agent-ready scientific infrastructure. Bohrium structures scientific execution around three capability domains---Reading, Computing, and Experiment---and provides a unified execution substrate that makes scientific assets callable, observable, and governable. Through consistent access interfaces for both agents and human users, these capabilities support reliable, tool-augmented scientific workflows spanning evidence access, computation, and experimentation.
• Figure 3: Workflow-oriented operation of SciMaster. SciMaster acts as a shared orchestration runtime that translates scientific objectives into executable, long-horizon workflows by coordinating agents and invoking Reading, Computing, and Experiment capabilities on Bohrium. Through explicit state management and execution traces, it supports inspectable, auditable, and reusable scientific work across a federated agent ecosystem.
• Figure 4: The emergence of a community-scale scientific flywheel. At platform scale (a multi-million scientific user base and large collections of tools, workflows, and models), scientific production can be organized by a coupled online--offline loop. The outer loop captures online execution on real tasks under realistic constraints and the resulting trace/validation signals, while the inner loop captures offline refinement (e.g., packaging, validation gates, routing policies, and model/knowledge updates) that feeds improvements back into subsequent execution.