JuliaHub has announced Dyad 3.0, a major release of its AI-native systems simulation platform for the design, refinement, and validation of complex physical systems. Dyad 3.0 introduces autonomous simulation agents that work alongside engineers to interpret specifications, mine prior designs and test data, generate candidate models, run physics-based simulations, enforce physical and safety constraints, and produce validated models and control code.
"AI has transformed software development through agents that combine LLMs with open-source compilers, but engineering physical systems requires the combination of LLMs with a physics compiler that grounds hardware designs in physical laws," said Dr. Viral B. Shah, CEO and co-founder of JuliaHub. "Dyad 3.0 brings agentic AI directly into the engineering workflow by combining autonomous agents, a multi-physics compiler, high-fidelity simulation, SciML, and enterprise deployment capabilities into one seamless environment. It gives engineers the leverage of AI while preserving the rigor, safety, and verification that physical systems demand."
Engineering teams designing aircraft, EVs, semiconductors, utilities, HVAC systems, medical devices, and other complex industrial systems face a common constraint: innovation cycles are increasingly fast, but validated engineering models remain difficult, manual, and time-intensive to build.
With Dyad 3.0, engineers can provide a requirements document, a prior generation design, historical test data, and a plain-language request. Dyad agents can assemble the model, explore thousands of variations, enforce physical and safety constraints, surface trade-offs in plain language, and produce validated code ready for hardware deployment.
The engineer remains the decision-maker-setting direction, evaluating trade-offs, and approving final designs-while Dyad automates the repetitive work of model construction, controller tuning, simulation execution, and toolchain integration.
AI adoption has accelerated in software development, but physical engineering has lagged because the work is governed by physics, safety, and verification requirements. General-purpose language models can assist with analysis or documentation, but they cannot reliably validate how a physical system will behave under real-world constraints.
Dyad was built to close that gap. By combining autonomous agents with physics-based simulation and Scientific Machine Learning (SciML), Dyad converts AI assistance into validated engineering productivity. The result is an AI-native workflow that can reason from requirements, simulation models, operational data, and prior designs, while enforcing physics throughout the process.
New in Dyad 3.0
The latest features include:
- Agentic model generation and design exploration: Agents interpret requirements, propose design candidates, run simulations, and refine models with engineer oversight.
- Digital twin workflows for predictive maintenance: Expanded platform capabilities help teams design and optimize industrial predictive maintenance applications.
- Agent-driven HVAC system design: Fast modeling tools, accurate refrigerant splines, expanded library coverage, and templates for common system architectures.
- FMU interoperability: Major Functional Mock-up Unit(FMU) advancements improve integration with the broader engineering toolchain.
- Multibody dynamics preview A preview expands Dyad toward robotics, vehicle dynamics, aerospace mechanisms, and other complex motion systems through 2026.
- Enterprise deployment readiness Improved installation, configuration, security, compliance, and lifecycle management support regulated and distributed engineering organizations.
Business impacts for engineering leaders:
Reduces manual model construction and iteration, lowering engineering hours per program and reducing late-stage prototype rework.
Shortens validated design cycles, enabling teams to pursue more programs with the same headcount and accelerate time-to-market.
Grounds design exploration in physics-based simulation; safety, regulatory, and operating constraints can be encoded and enforced across workflows.
Innovation speed allows teams to explore larger design spaces, multi-physics couplings, and what-if scenarios that are difficult to staff manually.
For more information contact:
JuliaHub
186 Alewife Brook Pwky., Ste. 212
Cambridge, MA 02138
617-925-6965
info@juliahub.com
www.juliahub.com