From one network to many sources

Silicon Network started as one thing: a network of Macs that serves open models behind one API. That is still the part that works today. What changed is the frame around it.

What changed

We now describe Silicon Network as a compute aggregator. Compute exists in many places: independent operators, decentralized networks, GPU clouds and the large public clouds. Each has its own console, its own prices and its own terms. Choosing between them means learning all of them.

So the site now starts from the workload instead of the source. In the chooser you pick what you want to run (AI inference, training, rendering, mining, trading bots or servers) and what matters most, price or speed, and it shows which kinds of source can run it. Each workload also has a page of its own, such as AI inference or rendering, that says what is offered today and what is planned.

What is live

AI inference through the Silicon Network API. It speaks the OpenAI, Claude Messages and Ollama formats, so most clients need only a new base URL and a key. Requests are served by operators' Macs running the Silicon Network engine on Apple Silicon.

The network is waiting for its first operators. No Mac has joined yet, so the API cannot serve a request until one does. The status page and the network map show this as it is: zeros, not estimates.

What is planned

Training, rendering, mining, trading bots and servers, and sources beyond our own operators. None of them is connected. The chooser says so for each one.

Some of the groundwork is built and switched off:

• The Linux node agent, which would let operators offer GPU rigs and servers for work other than inference. It would run each job in a locked down container. Started today, it registers nothing and runs nothing.
• The interface for outside providers. No provider is connected, so nothing from one is listed or sold.

There are no dates here on purpose. A source is listed when it is connected and checked, not before.

A monochrome identity

The visual identity changed with the positioning. It is black, white and the grays between them, with no accent color. The artwork is black stone with one band of white light through a gap: two independent forms and the connection between them. The primary line says the same in words: independent machines, one network.

Color is left for state only: an error, a warning, something live, and always with a label next to it. The site is still a Mac desktop, because the first workload still runs on Macs.

What to do now

• If you build with open models, read the quickstart and create a key. Requests are served once the first operators join.
• If you have a Mac with memory to spare, the operator guide explains how to join.
• If you need a workload that is not available yet, tell us what you need. Requests help decide which sources to connect first.