Browser node
Just open a tab. Nothing to install: your browser starts serving lightweight jobs.
Setup effort
Throughput
Real-time SOL payouts are live
Your GPU mines intelligence, not
Your GPU mines intelligence, not hashes
GRAID links idle graphics cards around the world into one global computer for AI. Your machine runs inference jobs, the network verifies the work, and you get paid on-chain.
$graid node start --wallet 7xKX…sAsU
GRAID network · demo
onlineinference
move your cursor over the grid
- 01Idle GPU
- 02Running node
- 03AI inference
- 04Verifiable receipt
- 05SOL payout
01 / 05
Your graphics card sits unused most of the day. GRAID puts that idle time to work.
How it works
From idle to income in five steps
This is not passive points farming. You earn only when your machine does real work and the network confirms it.
01
Open a node
A browser tab with WebGPU, or a native client running Ollama.
02
Join the pool
Your machine becomes part of a shared compute layer for open AI.
03
Receive jobs
The router sends inference requests to the best-fit nodes.
04
Pass verification
Every job closes with a verifiable receipt.
05
Get paid
Payouts land in your wallet on Solana.
75%
of every verified job goes to the node that did the work
$745
projected per month for an RTX 4090 at full load
3
large open models live on GRAID Pro
0
gatekeepers between your GPU and the market
Network principle
- No hyperscalers.
- No gatekeepers.
- No single point of control.
Every browser tab. Every native node. Every idle GPU. They all feed one shared compute layer, and every new node makes the network cheaper and more powerful.
◆ The compute exchange is live
Two sides. One global pool.
GRAID turns AI compute into a market that settles on Solana. One side buys inference, the other sells idle time on their graphics cards.
Buyers
Rent inference
- Top up in USDC and receive credits.
- Run open models through the playground or an OpenAI-compatible API.
- Pay per token, and only for completed responses.
- Every job comes with a verifiable receipt.
Buying requires a connected Solana wallet (Phantom, Solflare or Backpack). Signing in is free.
Providers
Sell your idle GPU
- Run a node in a browser tab with WebGPU, or natively with Ollama.
- Serve real inference demand from the shared pool.
- Stay online, build reputation and unlock higher-value jobs.
- Withdraw USDC on Solana.
Revenue split for verified work
Paid out in USDC for every confirmed jobKept by the protocol
Choose your node
Any computer can take part
Browser nodes make contributing open to everyone. Native operators bring more throughput, more jobs and higher earning potential.
Native node
Your card's full power on the network. More jobs, faster reputation, access to higher-value work.
Setup effort
Throughput
GRAID Pro
For large models that don't fit on browser or consumer nodes. Same gateway, same API.
Model size
Built forbuyers
Run a node
Join the network in a couple of minutes
All you need is a Solana wallet address for payouts, for example from Phantom. The node downloads a model, passes its check jobs and starts receiving work.
Option 1 · browser
Open a tab and press Start
- Open the node page in Chrome or Edge on a computer with a graphics card.
- Paste your Solana wallet address.
- Pick a model and press "Start node". Keep the tab open.
This button goes live on the GRAID network site once the router is deployed to a server.
Option 2 · Ollama
Native node at full power
- Install Ollama from ollama.com.
- Pull a model and start the GRAID node:
terminal
ollama pull llama3.1:8b
python graid_node.py \
--router https://your-graid-domain \
--wallet YOUR_SOLANA_ADDRESS
graid_node.py ships with the GRAID node package. Requires Python 3.8 or newer.
Earnings calculator
What your card could earn
An RTX 4090 at full utilization projects to about $745 a month. Move the sliders to see how uptime, network load and electricity prices change the result.
Projected net per month
$0/mo
Earned for verified work$0
Electricity−$0
Energy used0 kWh
Per day$0
The RTX 4090 reference comes from GRAID materials. Figures for other cards are rough estimates for illustration. Actual earnings depend on hardware, uptime, network demand and electricity costs, and are not guaranteed.
Market integrity
Paid for work that can be checked
Buyers see a receipt for every job. Providers are paid only for confirmed results. Three mechanisms keep both sides honest.
⬡ canary
Canary jobs
The network mixes in jobs with a known answer and immediately sees when a node answers wrong.
⬡⬡ redundancy
Redundancy checks
Some jobs run on several nodes at once, and the results are compared against each other.
⬡⬡⬡ reputation
Reputation scoring
Steady uptime and accurate answers raise a node's score and unlock higher-value jobs.
GRAID Pro is live
Large open models through one gateway
Pro adds a cloud-backed execution tier for models that don't fit in browser WebGPU or on a consumer card. Same OpenAI-compatible SubstraOS gateway.
Llama 3.3 70BPro
Qwen2.5-Coder 32BPro · code
Llama 3.1 8BFast
Request→Router→Browser/Native node/Pro
python · example
from openai import OpenAI # Same SDK, different gateway URL client = OpenAI( base_url="https://your-graid-domain/v1", api_key="GRAID_API_KEY", ) r = client.chat.completions.create( model="llama-3.3-70b", messages=[{"role": "user", "content": "Hello, GRAID"}], ) print(r.choices[0].message.content) # Billed per token, only for completed responses
$GRAID
Join the global AI computer
$GRAID is building a people-powered compute layer for artificial intelligence that anyone can join. Your compute powers someone else's AI, and its use pays you.