Get started
How it works

How it works

A request or a job comes in through one interface. Behind it are several kinds of source. Today one path is live: AI inference, served by operators' Macs.

From a request to a sourceA request or job goes to the Silicon Network interface, which checks the key and the balance and picks a source. Four kinds of source sit behind it: Silicon Network operators, decentralized networks, GPU clouds and public clouds. Only the path to operators, for AI inference, is live; the others are planned. On the inference network, answers can be verified and usage is metered and billed.
Request or jobAPI call, or a choice in the chooser
Silicon NetworkAPI and chooser: key, balance, routing
Inference network
Verification
Metering and billing
Silicon Network operatorsLive
Decentralized networksPlanned
GPU cloudsPlanned
Public cloudsPlanned
LivePlanned

Step by step

  1. You send a request or describe a job

    For AI inference that is an API call in the OpenAI, Claude Messages or Ollama format. For the other workloads you choose in the chooser; nothing can be ordered there yet.

  2. The interface admits it

    The gateway checks the key, its limits and the balance before any machine sees the request.

  3. Routing picks a source

    For inference, the gateway sends the request to an operator's Mac that has the model loaded. Planned sources would be picked by the priority you chose: price, speed, or both.

  4. The answer is checked

    On the inference network the gateway can re-run a sample of requests on a trusted node, send known prompts and compare the answers, and nodes sign what they serve. These checks are built and are switched on before operators outside the team serve. A connected provider's machines would be covered by that provider's own terms and checks.

  5. Usage is metered and billed

    The gateway counts the tokens of each request and the control plane charges them to your balance at the published rates. A signed receipt is available on request.

What is live

AI inference through the API is live, and serving starts when the first Macs join. No outside source is connected, and no other workload is available.

Choose computeRead the docsAbout Silicon Network