How it works

Sunlight in.
Answers out.

Node 01 is a shipping container in a remote location that runs AI models on solar power. No grid. No cloud GPU rental. When you send a question — through the chat page or an API call — it travels to that container, gets processed by a local model, and the answer comes back to you.

The power side

Where the watts come from.

Running AI models is power-hungry work. Node 01 generates and stores its own electricity so inference can keep going after sunset.

01

Sunlight hits the array

Solar panels on the container roof turn daylight into DC power. Standard charge controllers and inverters route it to the battery bank and the compute rack — no utility connection, no diesel generator.

02

Batteries store the day

Surplus solar charges a lithium battery bank. Stored energy covers cloudy hours, evening traffic, and overnight runs — so the cluster does not shut down when the sun goes down.

03

Hardware runs the models

Inside the container, a cluster of Ryzen AI Max+ 395 systems does the actual math. Filtered outside air flows through ducted intakes to keep chips cool without power-hungry compressors.

04

Wireless link to the internet

The site has no fiber run to the door. A point-to-point wireless antenna beams traffic across the valley to a relay with backhaul, so your chat or API request can reach the container and the answer can reach you.

The software side

What runs on Node 01.

Three layers sit between the internet and the model. The website and chat page are just clients — inference happens on the container.

Edge

Cloudflare Tunnel

Publishes our API at api.solarnodes.org and carries traffic into Node 01 over an encrypted tunnel, so the container never needs open inbound ports.

Gateway

LiteLLM

Speaks the OpenAI API format — chat completions, model lists, streaming. Verifies your API key and forwards each request to the right model.

Inference

Ollama

Loads open-weight models into memory on the Ryzen AI Max+ cluster and runs inference. Responses stream back through the gateway and tunnel to your client.

Available models

Pick a model ID in your API call or in the chat dropdown:

  • llama-8b — Llama 3.1, fast default
  • qwen-2.5-14b — Qwen 2.5 14B
  • qwen-3.6-27b — Qwen 3.6 27B
  • deepseek-r1-32b — DeepSeek R1 32B

Point any OpenAI client at our API base URL with your key. Endpoint reference, curl examples, and SDK snippets are in the documentation.

Open weights

Models you can run yourself.

An open-weight model is published as a file — billions of numbers that define how the AI behaves. Anyone can download those weights and run them on their own machine. Tutorial: run models locally

At SolarNodes we download each model once, load it through Ollama, and serve every request locally. Your prompts never reach Meta, Alibaba, or any other model maker's servers. There are no per-token fees and no usage data flowing back to the companies that originally trained the weights.

If you have used ChatGPT or the OpenAI API, the experience feels similar — but the model runs on our solar-powered hardware, not in a hyperscale data center.

The result

Solar-powered compute with an OpenAI-compatible API.