Network status: online

Rent the world's idle GPUs.
Or earn from yours.

One API for video, image, LLM and coding workloads — routed to a live mesh of real NVIDIA GPUs. No datacenters, no fixed servers. You pay per finished task; providers get paid for every idle cycle.

Interface
One API
Task types
4 · video·image·llm·code
Billing
Per task
Servers to run
0
[ Architecture ]

Three layers. One clean API in the middle.

Apps never touch hardware. Providers never touch your users. The orchestrator matches every task to a capable GPU, collects the result, and calls you back.

[ 01 · CONSUMERS ]

Vertical apps

Products like CinemaHub (video), and soon coding & chat assistants. They own the UX and billing, and send universal tasks to the API.

[ 02 · ORCHESTRATOR ]

gpusharing.org

Authenticates apps, queues tasks in Postgres, matches an agent by task_type and VRAM, stores the artifact, and fires your webhook.

[ 03 · PROVIDERS ]

The GPU farm

Desktop agents on real machines. One host, many workloads — ComfyUI for video/images, Ollama/vLLM for text & code.

[ Feature · universal_api ]

One interface, many workloads

Every job — a cinematic render, an image, an LLM completion, a coding task — is the same POST /tasks shape. Add a workload without touching the provider side.

[VID]
Video
comfyui_workflow
[IMG]
Image
sd_image
[LLM]
Text
llm_inference
[COD]
Code
soon
[ Module · provider_mesh ]

Earn from idle hardware

Join the mesh with one desktop app and get paid for every task your card finishes — no bidding, no lock-in.

For providers

Your GPU pays rent while you sleep.

Install one desktop app, tick the workloads you allow, and start earning for every task your card finishes.

  • One installer — pick Video & Images or Text & Coding (or both)
  • Only downloads the models for modes you enable
  • Runs quietly in the tray; pause anytime
  • Earn per finished task — no bidding, no lock-in
Download_the_host →
For developers

Compute as a single POST request.

Send a task, get a webhook. No GPU ops, no queue to run, no drivers to babysit.

POST /api/v1/tasks
# submit a task, receive a task_id, get called back when done
{
  "client_app_id": "cinemahub_prod_001",
  "task_type":     "comfyui_workflow",
  "required_vram_gb": 12,
  "priority":      "high",
  "payload":       { /* Wan 2.2 graph */ },
  "webhook_result_url": "https://api.cinemahub.ai/v1/result"
}
video · comfyui_workflow image · sd_image llm · llm_inference code · coming
[ Ecosystem ]

Built for one product. Ready for many.

The same network that renders cinematic video today will run your coding agent, chatbot and image tool tomorrow — without changing a line on the provider side.

First anchor client
Live in production
CinemaHub.ai
AI video studio
video · comfyui_workflow
Next up
chat · code · image
[ Join the network ]

Idle silicon, meet demand.

Ship AI features without buying GPUs — or turn the card you already own into an income stream.