What is c0mpute?
c0mpute is a decentralized AI inference network. Instead of routing your prompts through corporate data centers, c0mpute connects you directly to a distributed network of GPU workers — regular people sharing their compute power.
AI powered by people, not data centers.
How it works
You send a message. The orchestrator finds an available worker. The worker runs the model on their GPU and streams tokens back to you in real-time. No middleman logging your prompts. No corporate filter deciding what you're allowed to ask.
Three tiers
| Tier | Model | Where it runs | Notes |
|---|---|---|---|
| Free | Qwen 1.5B | Browser (WebGPU) | ~900MB, fast, basic |
| Pro | Dolphin Mistral 7B | Browser (WebGPU) | ~4GB VRAM, uncensored |
| Max | Qwen2.5 14B abliterated | Native workers | ~9GB, uncensored + web search |
The stack
- Browser workers use WebGPU via WebLLM to run models directly in the browser tab
- Native workers use node-llama-cpp with CUDA, Metal, or Vulkan acceleration
- The orchestrator is a Socket.io server that handles job routing, worker matching, and real-time token streaming
- Workers earn SOL for every job they complete
Why?
Centralized AI providers censor their models, log your prompts, and can revoke access at any time. c0mpute is the alternative — private, uncensored, and owned by no one.
Anyone can use c0mpute or contribute compute and start earning.