Curlscape AgentCrew
Deploy AI agents that handle the repetitive layers of simulation work — geometry preparation, meshing, solver setup — so your engineers spend their time on actual engineering. Launch parallel agents with planning and end-to-end orchestration, and multiply the throughput of your CAE team.
- Pick it when
- Your team runs many simulations and loses its days to setup.
- How you use it
- Describe the run in a sentence. The agents plan it, run it, and you supervise.
- Ends with
- A solved case, checked at each stage, and an engineering report.
- Works in
- Siemens NX
- Solid Edge
- FreeCAD
- Ansys meshing and solvers
- snappyHexMesh
- OpenFOAM
AgentCrew screen recording: an engineer types a one-line request, the agents work through material assignment, physics, boundary conditions and solver setup with a check mark per step, and the run ends with a 3D temperature field and a folder of reports and plots.
The agent lineup:
- Planning agents
- Complex simulations demand upfront planning across physics, assumptions, prior comparable runs, and test data. Planning agents assimilate this context through smart connectors, produce a structured simulation plan, and execute it autonomously — or iterate on it with you.
- Geometry prep agents
- Clean, simulation-ready geometry (watertight for CFD; defeatured and well-connected for FEA) is the prerequisite for reliable results. Geometry prep agents automate cleanup with both the analyst's simulation intent and the CAD engineer's tool knowledge built in.
- Meshing agents
- Encode meshing best practices and iterate autonomously against your target quality metrics — running until the mesh meets spec.
- Solver setup agents
- Wrong boundary conditions, operating conditions, and physics setup sink a significant share of simulations. Setup agents encode experienced-engineer judgment so simulations launch right the first time.
- Reporting agents
- Engineering reports generated automatically, following your formatting standards and required content.
Connectors — how AgentCrew works with your stack
An agent is only as useful as what it can read and act on. Connectors are how AgentCrew plugs into your existing toolchain — no rip-and-replace, no vendor lock-in.
Every connector does two jobs:
- Read context
- Pull in the material an agent needs to plan and decide — geometry and model files, simulation setups, past runs, mesh quality reports, test data.
- Execute actions
- Drive the tool itself through its scripting and API layer — running cleanup operations, generating meshes, configuring solvers, launching runs.
Current coverage:
| Domain | Commercial | Open source |
|---|---|---|
| CAD |
|
|
| Meshing |
|
|
| Solvers |
|
|
Don't see your tool? Connectors are built to demand. If your stack includes a tool we don't cover yet, we scope the connector as part of your pilot — most take days to weeks, not months, because the agent architecture is tool-agnostic by design.
FAQ
Clearly answered
01What does Curlscape AgentCrew automate?
Curlscape AgentCrew automates the repetitive layers of simulation work with planning agents, geometry prep agents, meshing agents, solver setup agents and reporting agents. Planning agents can run a simulation plan end to end or iterate on it with you, and parallel agents multiply the throughput of a CAE team.
02Which tools does AgentCrew connect to?
Curlscape AgentCrew connects to Siemens NX, Solid Edge and FreeCAD for CAD, Fluent Meshing, Ansys Meshing and snappyHexMesh for meshing, and the Ansys solver suite and OpenFOAM for solving. Every connector both reads context, such as model files and past runs, and executes actions through the tool's own scripting and API layer.
See the coverage table03What if AgentCrew doesn't support our tool yet?
Curlscape builds the connector. New Curlscape AgentCrew connectors are scoped as part of your pilot, and most take days to weeks, not months, because the agent architecture is tool-agnostic by design.
04How does AgentCrew catch its own mistakes?
Review agents gate each stage of a Curlscape AgentCrew pipeline: geometry review, mesh QA, solver QA and run monitoring each have a defined check to pass before the pipeline advances. In one of our own studies, a verification agent caught a sign-convention error in post-processing before it reached the report.
Read what the agent caughtHow we run this ourselves
- Lab01
Multi-fidelity optimization of a propeller design space
2,560 propeller designs screened with fast blade-element theory, then corrected with 12 blade-resolved CFD runs, instead of running CFD on every candidate.
- Lab02
The sign-convention error the agent caught itself
A post-processing sign error slipped into an automated study. A verification agent flagged it as physically implausible before it reached the report.