AgentCAE

AgentCAE is the neutral contract layer connecting AI-native engineering products. AgentCFD and AgentFEM remain independent numerical platforms; AgentCAE gives them stable names for results, fields, coupling interfaces, scientific samples, provenance, and trust.

AgentCFD / OpenFOAM ─┐
                     ├─ AgentCAE contracts ─ coupling / datasets / Physics AI
AgentFEM / FEniCSx ──┘

The package is intentionally small and dependency-free. It does not contain a CFD solver, a finite-element solver, a neural-network framework, or a second modeling language.

First release

The first pre-alpha release freezes the existing agentcae.* identities already emitted by AgentCFD:

• agentcae.simulation-result/0.1.0;
• agentcae.scientific-sample/0.1.0;
• agentcae.field-bundle/0.1.0;
• the legacy agentcae.coupling-manifest/0.1 identity.

It provides dependency-free discovery and fail-closed header validation while the complete JSON schemas and two-way adapters are promoted from exercised AgentCFD and AgentFEM implementations.

from agentcae import require

require(record, "scientific-sample")

agentcae doctor --json
agentcae contracts --json
agentcae validate scientific-sample sample.json --json

Architecture policy

• Products share contracts, not solver internals.
• A successful solve is not automatically trusted engineering evidence.
• Field location, units, components, mesh identity, time coordinates, and processing history remain explicit.
• Conservative interface transfer and physical acceptance are never inferred from file compatibility alone.
• Optional coupling and learning providers remain separately installable.

AgentCAE is pre-alpha. Version 0.1.0a1 establishes package ownership and a small compatibility surface; it does not claim completed CFD/FEM coupling.

License

Apache-2.0.