What is GPU Mesh?
A free, community-powered network for using AI language models. GPU owners share their spare computing power — you use it through tools you already know. No credit card, no new apps.
In plain language
How it works
You ── HTTP ──▶ Coordinator ── WebSocket ──▶ Donor's GPU ──▶ AI Model
1
Share
GPU owners run a small agent program. It connects their hardware to the network.
2
Match
When you send a request, the coordinator finds an available GPU with the model you need.
3
Use
Your request runs on the donor's GPU. The response streams back to your tool. Nothing is stored.
For consumers
You want free access to AI models for coding, writing, or research.
You use tools like Continue.dev, Aider, Cline, or any OpenAI-compatible app.
You don't have a powerful GPU — or you do, but you want free inference.
For donors
You have a GPU that sits idle most of the time.
You run Ollama with some models.
You want to contribute to the community, earn reputation badges, and see your hardware put to good use.
Key facts
- Free — no payments, no credits, no catch
- Open source — MIT licensed, anyone can audit or self-host
- Private — your prompts and responses are never stored
- Community-driven — powered by volunteers sharing spare GPU time
- Standard API — works with any OpenAI-compatible tool; no new client to install
FAQ
Is it really free?
Yes. Donors volunteer their GPU time. There is no payment system, no credits, no "pro tier." It's free by design.
What models are available?
Whatever the community is running — typically Llama, Mistral, CodeLlama, Gemma, and other open-weight models via Ollama. Check /models for the current list.
Is it fast?
It depends on the donor's hardware and network. Consumer-grade GPUs over home internet won't match a datacenter. It's best-effort community infrastructure, not a paid SLA. For chat and coding, it's usually fast enough.
Can donors see my prompts?
Yes. Donors run the AI model on their own hardware and could inspect traffic. This is disclosed transparently. Don't send passwords, API keys, or other sensitive information.
How do I contribute my GPU?
Can I run my own coordinator?
Yes. The whole project is open source (MIT). Clone the repo, build the coordinator binary, and run your own mesh — private or public.