onnx
local neural inference runtime in cyb. runs ONNX models on device via burn-webnn — dispatches to NPU (CoreML, NNAPI, WebNN), GPU (cyb/wgpu fallback), or CPU depending on hardware
provides local AI without cloud dependency:
- SLM (~100-300M params) for legacy CSS bridge in cyb/features
- embedding models for local semantic search in cyb/oracle
- classification models for content type detection
- custom models loaded by cyb/robot progs
three execution targets:
| platform | backend |
|---|---|
| browser | WebNN API → NPU/GPU |
| desktop | burn → CoreML (macOS) / DirectML (Windows) / CPU |
| mobile | burn → CoreML (iOS) / NNAPI (Android) |
see cyb/features for how inference integrates with the render and compatibility layers. see cyb/wgpu for GPU compute fallback