ADR-0002 CompliantLocal Neural AI & APIs

Local Neural AI Assistant Engine

Physics-constrained ONNX neural inference engine running 100% locally on workstation GPU/NPU under ADR-0002 air-gapped security.

Inference Backend

onnxruntime-web / WebGPU / WASM

Execution Provider

Local Workstation GPU

Network Dependency

0 Bytes (100% Air-Gapped)

Safety Enforcer

2.5x Von Mises Yield Boundary

The ZylForge Local Neural Assistant parses natural language engineering intents and converts them into validated 3D parametric sketches.

Unlike public cloud LLMs, all neural model weights run in-process via ONNX runtime, ensuring sensitive defense IP never leaves local RAM.

Neural suggestions are automatically evaluated against the physical B-Rep kernel before rendering to prevent non-physical geometry.

Technical Code Example

Production API

// Local Neural Intent Query
const assistant = new ZylforgeLocalAIWorker();
const result = await assistant.processIntent("Optimize servo bracket for 500N lateral load");
console.log(result.insights); // Yields Tet10 mesh stress predictions

Documentation verified against ZylForge Release v1.0.0-rc1

Get Desktop Build