Install the host CLI
One command on macOS, Linux, or Windows (WSL). The CLI handles registration, NVIDIA driver checks, and a guided test run.
Compute marketplace
Sell what your GPU isn't using. Reclaim it whenever you want.
Connect a capable GPU outbound — no public address, no inbound ports. Apps occupy clear lanes on it; you earn while it's idle and reclaim it any time. Per-second metering means short bursts are still worth it.
How it works
Three steps to first dollar.
Verify with Stripe Connect
Standard KYC with our identity partner. Payouts arrive weekly to your bank, in your local currency where Stripe supports it.
Set caps and start earning
Decide how much VRAM and what time of day you offer. Reclaim the GPU any time — your work always comes first.
Earnings
What hosts typically take home.
Per-GPU ranges below assume 12–18 hours of online time per day at the current marketplace rate. Actual earnings vary with demand, geography, and uptime.
| GPU | Typical earnings | Lane |
|---|---|---|
| RTX 4090 (24 GB) | $12–28 / day | General purpose |
| RTX 5090 (32 GB) | $22–48 / day | General purpose / Reasoning |
| L40S (48 GB) | $36–72 / day | Reasoning / Frontier |
| H100 PCIe (80 GB) | $72–144 / day | Frontier reasoning |
| MI300X (192 GB) | $96–192 / day | Frontier reasoning (long context) |
FAQ
Common questions
No. The host CLI opens an outbound WebSocket to api.perchy.ai. It works on residential ISPs and behind most firewalls.
Yes. Set a VRAM cap (e.g. --max-vram 80%) and the runtime will leave headroom. You can pause earning instantly with `perchy host pause`.
Only the Perchy host runtime. Customer prompts and responses are processed in volatile memory and never written to disk. Model weights are streamed from the upstream provider where licensing requires it.
Weekly via Stripe Connect, after KYC clears. Minimum payout is $25; balances below the minimum roll over.
Ready to flip the switch?
Sign in, link payouts, and run npx perchy host start --earn.