Platform · Simulation
Simulation agents
Coupled thermal-stress simulation answers the hardest reliability questions in modern hardware, and it is still where the slowest, most expert-bound setup work lives. Agentriq turns a plain-language spec into a valid solver model tree and drives the solver stage by stage.
01The problem
Consider advanced electronics packages, from FCBGA to 2.5D and 3D-IC stacks, where thermal-mechanical warpage and stress decide whether a part survives assembly and field life: predicting them means coupled thermal-stress FEA, and industry tools such as Cadence Celsius and Synopsys RedHawk-SC Electrothermal show the scale of it. Even with a massively parallel solver running up to 10x faster than legacy tools, an expert still spends days assembling the model tree, materials, load steps, and boundary conditions correctly. The bottleneck is no longer compute, it is setup.
02How we solve it
Natural language becomes a valid solver model.
Agentriq reads an engineer's plain-language spec, assembles a physically valid model tree, and calls the solver stage by stage through typed APIs.
Natural-language to model setup
A written spec is parsed into geometry, materials, physics couplings, load steps, and boundary conditions, the way MooseAgent maps a described problem to valid MOOSE input cards.
Solver tool-calling via typed APIs
Agents drive COMSOL LiveLink, OpenFOAM, and Ansys through typed interfaces, so every stage is a validated call rather than brittle hand-edited input text.
Multi-agent simulation pipelines
Specialized agents for meshing, input writing, running, and review hand off in sequence, mirroring the six-agent OpenFOAM workflow published as Foam-Agent.
Knowledge-graph-augmented setup
A retrieval layer over solver input cards and documentation grounds each choice, as in the Neo4j GraphRAG of PDE-Agents and the hierarchical RAG index of Foam-Agent.
03What it produces
Valid models, far fewer retries.
A complete, runnable model tree from a single spec: in the published Foam-Agent benchmark this multi-agent approach reached an 88.2% end-to-end success rate across 110 OpenFOAM cases, up from 55.5% for the prior framework.
04The agentic loop
Setup, run, verify, decide.
Agentriq wraps solver setup in a closed loop that runs the model, checks the result against physics, traces root cause when it fails, and only accepts a run a judge has passed.
- 01Read the spec and assemble the solver model tree: geometry, materials, physics couplings, and load steps.
- 02Call the solver stage by stage through typed APIs, meshing and running each stage as a validated step.
- 03Verify outputs against physical checks: energy balance, convergence, and warpage and stress within expected bounds.
- 04On failure, trace root cause with Agentic RAG over solver documentation and prior cases, then repair the setup.
- 05A judge decides whether the run passes, and the verified model, results, and rationale are written back to the record.
05Tooling
Simulation, solvers & frameworks.
Get started