Browser
ONNX → WASM. The same graph executes client-side, no install.
ONNX model runtime · distribution
Models as artifacts.
ONNX in, verifiable out.
A runtime and distribution surface for machine-learning models. Models ship as content-addressed ONNX artifacts and run the same graph in the browser (WASM), on device, or on the server — with a content identifier you can verify before you load.
Registry
| Model | Format | Params | Runtime | CID |
|---|---|---|---|---|
| — | ONNX | — | wasm · native · server | bafy… |
| — | ONNX | — | wasm · native · server | bafy… |
Registry is empty pending first publications. Each row resolves to a model card, the ONNX artifact, and its CID.
Runtime targets
Native
On-device inference via the platform’s accelerators.
Server
Headless inference behind a verifiable boundary.
Verifiable
Content-addressed: verify the CID before the weights load.
Format & governance
Part of the .store family. The same compute core runs across surfaces; type identifiers and conformance are governed alongside standards.agency and utxo.store.