Rescale Puts Engineering Agents to Work on Hermes 2 Reactor Design

Rescale on November 10, 2026 introduced a framework for customizing engineering AI agents and disclosed that Kairos Power and Oak Ridge National Laboratory’s Manufacturing Demonstration Facility are using its platform for Hermes 2, a 50-megawatt fluoride-salt-cooled reactor.

Rescale AI developer engineers supported the collaboration, which is testing advanced manufacturing and construction methods. The teams used agents for thermal, flow and shielding simulations and associated reporting on reactor-cavity and bioshield-plug designs, with an engineer approving every agent action.

ORNL is encoding its simulation and additive-manufacturing expertise in agents so engineers beyond the facility can adopt those methods. The efficiency target is prospective: Kairos Power projects that agentic workflows built from past simulations could lift end-to-end simulation analysis efficiency by up to 3x while preserving the traceability needed for regulatory compliance.

Reusable Projects and Traceable Work

The Fall 2026 Showcase account makes Agentic Projects the control layer for Edison, Rescale’s AI assistant and agent orchestrator. Teams can set objectives, tools, skills and guardrails, then add agent nodes for steps such as validating inputs, analyzing results or preparing evidence. Projects can be saved and shared across programs, and a suggested learning enters project memory only after an engineer accepts it.

Engineering Intelligence links requirements, design, simulation, test and process records in knowledge graphs, giving agents the relationships and configurations needed for complex work. A separate work graph maps people, data and tasks across a workflow so engineers can trace how the work was performed and where agents participated.

The November 10 event was not the first appearance of every component. Rescale’s Agentic Projects release note is dated October 26, 2026 and describes validated baseline jobs, linked standards and files, reusable skills, and citations when Edison answers from a linked document. Projects begin private, can be shared with a workspace, and require Agentic Engineering to be enabled.

Desktop and External Client Boundaries

Rescale Desktop followed in a November 5, 2026 release note. The app for macOS on Apple silicon, Windows x64 and Linux x64 moves files between local and cloud environments, tracks jobs and provides Edison chat without a browser. Edison can search only user-approved folders; it can see file names, sizes and dates but not file contents, and asks before every transfer. Chat requires Advanced Modeling & Simulation or Agentic Engineering and is not yet available on Rescale ITAR.

Rescale also expanded its Model Context Protocol (MCP) connectors, which let an outside AI client issue natural-language requests to the platform. The company says the connectors make its specialized agents and more than 1,350 pre-installed engineering applications available in Claude and other MCP-compatible tools. Rescale MCP is also listed in the Claude and Codex directories.

The MCP documentation limits that route to interactive use by a person working through an agent. Each connection is bound to one workspace and the signed-in user’s existing permissions, with actions attributable to that user; production integrations are directed to Rescale’s REST API or command-line interface.

AI Physics for Surrogate Models

The AI physics additions address surrogate models, AI approximations trained from simulation data. A dataset-creation agent turns existing simulation data into training-ready datasets. Guided pre-processing identifies outliers and coverage gaps and recommends data splits, while Auto Tune adjusts model configurations during training.

Rescale says Design Explorer provides a guided local interface for evaluating thousands of design options against constraints set by simulation engineers and for linking multiple surrogate models into multiphysics optimizations. Uncertainty quantification adds confidence and out-of-distribution signals to each prediction, identifying results that need a full-fidelity simulation check and cases where the model needs expansion.

The release applies a consistent boundary: agent-suggested learnings require approval, Hermes 2 agents act with human sign-off, and uncertain surrogate predictions are flagged for full-fidelity checks. The nuclear collaboration therefore tests whether reusable agent workflows can reduce analysis effort while preserving an auditable engineering record.