One compatible API links AI builders to the compute network.

Build against a familiar request shape

The gateway design uses OpenAI-compatible endpoints for chat completions, embeddings and images. Builders select a compute lane for the request instead of assuming all GPUs offer the same privacy. A completed job has a receipt that can be retrieved by its job ID.

Choose the request’s compute lane

The X-Spindle-Lane header selects open; sealed in testing. The sealed lane requires eligible confidential GPUs and current attestation. The open lane serves non-sensitive requests on suitable hardware, including consumer GPUs. The selection is a privacy and hardware decision, not simply a visual setting.

Connect payment to the job

The buyer payment design supports a USDG API balance or per-call x402 settlement. The scheduler sends accepted jobs to suitable nodes, and metering records verified work in USDG. Production access details and launch information are not represented by this illustrative website.

Retrieve the work record

The receipt endpoint links the builder’s completed request to its signed COSE work record. A receipt can support execution and billing checks without asserting that an answer is correct. Providers then follow the hourly cumulative root and their chosen claim route.