Hey HF community,
I built a tool that might be relevant to a lot of people here who either have idle GPUs or need compute for quick jobs. It’s called RunSnack — share your GPU with a single link. No accounts, no cloud, no platform fees.
Why this matters for ML people:
Need an A100 for 2 hours of inference testing? Find a host, get a link, instant terminal.
Got a 4090 sitting idle overnight? Share it, set your price, negotiate directly.
No KYC, no cloud markup, no usage logs. Just P2P compute.
How it works:
Host runs a script → gets a shareable link
Renter opens link in browser → instant xterm.js terminal with GPU access
Everything runs inside a sandboxed Docker container (read-only, privilege-dropped)
WebRTC data channel — encrypted, P2P, no central server sees your data
Marketplace is organic: post on Discord what you need or offer, negotiate, share link
Built for the “quick GPU snack” use case:
ML researcher needs 30 min of inference on a GPU they don’t own
Student wants to fine-tune a model for a few hours without spinning up a cloud instance
Someone with a homelab GPU wants to monetize idle time
I’d love feedback from this community on:
Would you use this? Why or why not?
Security concerns? Docker sandbox + GPU passthrough — what would make you trust it enough to either host or rent?
What’s missing? Pre-installed ML frameworks? Persistent storage? Something else?
Pricing expectations? What would you pay per hour for, say, a 4090 vs an Jetson Nano via a P2P link like this?
Is this approach a competitor for sites like RunPod?
Thank you!
Lastly, I add this 30-sec YouTube reel so everyone gets better the idea