For buyers
Rent a GPU by the hour. Use it from your laptop.
Top up your prepaid balance, pick a workload, autopilot books you a GPU and gives you a private OpenAI-compatible URL. Pay per rental, not per token. Cancel mid-stream for a pro-rated refund.
How it works
Five steps from signup to your first AI response. Takes about three minutes.
- 1
Sign up + top up
Email + password. No card required to sign up. When you're ready, add credit via Stripe ($25 minimum, you pay the standard 2.9% + $0.30 processing fee — we don't mark it up). Balance lives in your account, debited per rental.
- 2
Pick a workload from the dashboard
Four cards on /dashboard: 💬 Chat · 🎨 Image · 🎙 Transcribe · 🔎 Search/Embeddings. Click one — Wattlend's autopilot finds the cheapest matching GPU pool and shows you the price before booking.
- 3
Get a private URL + access token
Each rental gets a unique URL (https://wattlend.com/v1/rentals/<id>/chat/completions) and a Bearer token. The URL is yours alone — no other buyer can hit it.
- 4
Use any OpenAI-compatible client
Drop the URL + token into the OpenAI SDK (change two lines), Open WebUI, LibreChat, Cursor, Continue, Chatbox, or anything else that speaks the OpenAI API. Or use the in-Wattlend chat UI on the rental detail page.
- 5
Pay by the hour, cancel any time
Rentals are billed by the hour with the rate locked at booking. Cancel mid-rental for a pro-rated refund on unused time. Refunds land in your balance immediately.
Works with what you already use
Wattlend speaks the OpenAI API. Every rental's detail page has step-by-step setup instructions for the clients below — same Bearer token, same URL pattern, no proprietary SDK.
What you can run
Llama-3, Mistral, or whatever the seller has installed via Ollama. OpenAI-compatible chat completions endpoint. Streaming, system prompts, multi-turn — all standard.
SDXL out of the box, 1024×1024 by default. POST a prompt, get an image URL back. ComfyUI pre-installed on hosts that advertise it.
Whisper-large-v3. Upload audio, get text back with timestamps. Runs on CPU-only sellers too, so it's the cheapest workload tier.
BGE-base embeddings (768-dim). POST a batch of text, get vectors. Cheap and fast — most users embed an entire knowledge base in under 10 minutes.
Llama-3 LoRA on your dataset. Provide a HuggingFace dataset URL or upload JSONL. Adapter saved to a returnable URL when done. Auto-checkpoints every 15 min so a seller dropout doesn't kill the run.
Power user only. POST a Docker image reference; the seller's agent pulls and runs it with the GPU you reserved. For anything Wattlend doesn't have a template for.
Try it free — no card required to sign up
Create an account in 30 seconds. Top up when you're ready to send your first request.