You bought hardware that sits idle most days. You rent GPU time you only half-use. List it on the TokenFlow marketplace and earn per million tokens served. We send the traffic, you serve the model, we split the revenue.
No vetting process, no application form, no minimums. Sign up, register your hardware or your endpoint, set the price you want, and you're in the pool. The first paying request can land within minutes of you coming online.
Free account. No card needed to list. You'll need one later to receive payouts.
One form. Self-hosted GPU? Cloud endpoint? Both work. Pick which.
Minimum dollars per million tokens. The marketplace routes work to whoever's cheapest at the quality bar — that includes you.
Earnings accrue in real time. Withdraw any time you want.
RTX 4090 in your living room. Old miner you turned into an inference rig. Three nodes in your garage. Run our small agent — it forwards work from the gateway to whatever model server you're already running, locally. We never need access to your machine.
Works with any local OpenAI-compatible model server. Set the agent up once, leave it running.
RunPod serverless, Modal, Together AI, Fireworks, Groq, your own production API behind a domain. Paste in the URL and the API key it expects. We dial it directly when work routes to you. No agent required.
Works with anything that speaks the OpenAI /v1/chat/completions shape.
Simple split. The customer paid $X. We take 20%. You keep 80%. No setup fee, no monthly minimum, no per-request listing charge.
Earnings accrue per request and post to your account in real time. Withdraw any time after the first $20 — same-day for crypto, 1–3 business days for bank transfer.
For each node, you pick a minimum price you'll accept per million tokens. We never route work to you below that floor. Above the floor, the marketplace picks among available providers based on price and quality — so the cheaper your node, the more work you'll get, but you'll never be forced to take below your minimum.
Raise or lower your minimum any time. Changes take effect on the next routing decision — usually within seconds.
Need your GPU back for something else? Toggle "accept work" off in your dashboard. Routing stops immediately. Toggle back on when ready.
Decide it's not worth it? Delete the node from your dashboard. Your earnings stay; only future routing stops.
If your hardware can't compete with cloud GPU prices, you won't get traffic. The math works best for: idle high-end consumer GPUs (4090s, etc.), spare capacity on rentals you're already paying for, and cheap-region cloud endpoints.
No commitment. List a node, earn $30 in a weekend, unlist Monday. Or run for years. Up to you.
The agent makes outbound connections to us. Nothing inbound. We never SSH in, never need root, never see anything but the model requests we route through. Your machine, your rules.
Customers pay our retail rate. You earn from their payment. We don't subsidize either side. If nobody's buying inference at your price floor, you don't get routed.
https://api.example.com/v1)The hardware is yours. The price is yours. The customers come to us.
List your first node