Compute
from everywhere.
BRAIN routes every request to the cheapest path that can run it, browser compute, cloud GPUs or external models, and attaches a receipt. Your computer can power it and earn credits, USDC or SOL.
One request, traced.
From API call to paid contributors. Scroll to step through it.example trace
- 01Request
- 02Gateway
- 03Router
- 04Split
- 05Nodes
- 06Verify
- 07Merge
- 08Response
A developer sends one request
POST /v1/chat/completions with model brain/auto. Same shape as the OpenAI API, so existing SDKs work unchanged.
A developer sends one request
POST /v1/chat/completions with model brain/auto. Same shape as the OpenAI API, so existing SDKs work unchanged.
- 0.000sPOST /v1/chat/completions model=brain/auto priority=cheap
- ▍
One datacenter GPU.
The usual way to serve AI: a single 80 GB accelerator in a rack, rented by the hour.
Illustration of the device-class floorplan · counters above are real
Live topology
Four classes of supply, each shown in its real state. The graph below is the browser network: requests enter, split across device classes, execute, and merge. Green markers are real browsers connected right now; grey cells are simulated scale.
Your computer
can power BRAIN.
No install, no driver, no CLI. Open a tab, let it measure your GPU, and join. The server issues a challenge only a real GPU can answer in time, then starts sending verifiable work. Earn credits, USDC or SOL.
- 01DetectReads exactly what WebGPU exposes. Anything it can’t see is labeled unavailable, never guessed.
- 02BenchmarkA WGSL kernel on your GPU. Score comes from the server’s clock, not yours.
- 03JoinYour node appears in the topology and starts receiving jobs within seconds.
What your
GPU earns.
Rewards are paid for verified compute, from creator fees and inference sales. Holding tokens raises your multiplier, up to a cap. Holding alone earns nothing.
The contributor pool is funded by creator fees paid to the protocol wallet and by inference sales. Nothing has been recorded yet, so the real pool today is $0 and any dollar figure here would be made up. Measuring your GPU on /earn is real and takes about a minute. The estimator can run against a modelled network, clearly labelled SIM.
One chat.
Every receipt.
Ask in the app or call the OpenAI-compatible API. BRAIN AUTO estimates every execution target, picks one for your mode and privacy, and tells you exactly where the answer ran and what it cost.
- brain/autoAutomatically chooses the cheapest execution path that satisfies the request.
- brain/qwenDistributed Qwen-class inference across the browser pool.
- brain/codeCoding-optimized inference on distilled reasoning models.
- brain/embedDistributed embeddings. Small model, highly parallel, browser-native.
curl https://brainnetwork.app/v1/chat/completions \
-H "Authorization: Bearer $BRAIN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "brain/auto",
"messages": [{"role": "user", "content": "Explain this contract"}]
}'Drop-in for OpenAI SDKs: set base_url to your BRAIN endpoint. Examples →