ShareMyAI PRIVATE ALPHA Open console

OPEN INFERENCE / FROM ANY CAPABLE BROWSER

The browser is
the network.

Request AI work or share local compute. ShareMyAI routes open models to qualified browsers, verifies every output, and settles useful work on-chain.

SUPPLY COMPUTE Bring this browser online Qualify WebGPU, install the pinned model, and run useful work locally. REQUEST WORK Create with the network Submit an image request and watch it move from escrow to verified output.

Private alpha · no-value test token · experimental runtime · transparent job state

ONE REQUEST / VISIBLE END TO END

No black box.
Just a clear path.

  1. 01
    QueuePrompt + reward enter escrow
  2. 02
    LeaseQualified browser claims work
  3. 03
    WebGPUPinned model renders locally
  4. 04
    QuorumTwo independent output votes
  5. 05
    SettlementEscrow releases SMAI-TEST

USE THE NETWORK

Create something.
Or make it possible.

One product for both sides: people who need inference and people with capable hardware ready to run it.

REQUEST

Describe the result. The network handles the route.

Connect once, fund a request, and follow its path from queue to verified image without leaving the workspace.

Open a request →
CONTRIBUTE

Qualify once. Put local silicon to useful work.

A canonical render proves the browser and model together. Then this device can claim real requests from the shared queue.

Bring this browser online →

TRUST, MADE VISIBLE

Useful compute.
Verified by design.
Still an alpha.

01

Local by default

Device checks and model execution happen in your browser. Hardware details are only shared if you explicitly submit minimized cohort evidence.

02

Guarded settlement

Jobs use an ERC-20 test token and escrow contract on Robinhood Testnet. SMAI-TEST has no monetary value and is not a production token.

03

Measurable before scalable

The first goal is repeatable output across real devices—not token speculation, uptime theater, or unverifiable mining claims.

ShareMyAI Compute workspace
Network online TESTNET

WALLET PROVIDER

Choose a wallet

Select the extension that should authorize this Robinhood Testnet session.

YOUR COMPUTE WORKSPACE

What do you want
to do today?

Request open-model inference or bring this browser online. Every step stays visible from wallet to verified result.

REQUEST WORKCreate an image→ SUPPLY COMPUTEBring this browser online↗
LIVE PACKET PATHCONNECTING
  1. QUEUE— open
  2. LEASEbrowser claim
  3. WEBGPUlocal inference
  4. QUORUM2-of-N verify
  5. SETTLEtestnet escrow

DEVICE & MODEL / LOCAL QUALIFICATION

Bring this browser
online.

Prove this browser can reproduce the network workload. The check is local, the model stays on-device, and qualification only activates after verifier quorum.

NODE READINESS

Four checks. One trusted worker.

A device is only “qualified” after a canonical model render—not from browser claims alone.

  1. 01 GPU access Adapter + limits
  2. 02 Shader proxy Measured matmul
  3. 03 Model manifest Klein Q4 contract
  4. 04 Canonical render Runtime adapter needed

MEASUREMENT

Device telemetry

LOCAL ONLY
Compute proxy — GFLOP/s · F32 matmul
Storage binding — largest shader buffer
Shader F16 — feature availability
Model runtime — adapter contract
REPORTED ADAPTER

Waiting for permission

BROWSER

Detecting…

TARGET WORKLOAD

FLUX.2 Klein 4B

APACHE-2.0 MODEL
4steps
Q4weights
256²alpha target
Manifest not checked

The benchmark will validate the model-pack contract. Weights are not downloaded in this spike.

LOCAL MODEL PACK 5.23 GB

Run the device check before installing the pinned model pack.

Experimental runtime and download terms

Installation downloads 5.23 GB from a pinned community ONNX conversion into this browser's Cache Storage. The ShareMyAI adapter is independently implemented from upstream Apache-2.0 sampling code. Converted artifact licensing and production integrity review remain launch gates; this is closed-alpha software.

RECEIPT PREVIEW

Run log

NO RECEIPT
  1. Ready. Start a device check when you are ready.

OPT-IN COMPATIBILITY EVIDENCE

Count this canonical attempt

Shares: coarse GPU class, bucketed limits, benchmark/render timing, success or failure, fixed workload/runtime IDs, and output hash. Never shares your prompt, image, wallet, participant ID, user agent, adapter description, or event log.

RUNS
— / 30
COMPLETE
—
DEVICE CLASSES
— / 3
PAID JOBS
— / 10

No data is submitted automatically. Complete or stop one canonical render to opt in.

SHARED WORKSPACE

Request work. Run work.

CONNECTING ROBINHOOD TESTNET · NO-VALUE TOKEN
REQUEST

Create an image

0 test credits
PARTICIPANT creating…

Need transaction gas? Get Robinhood Testnet ETH from Alchemy. Testnet ETH and SMAI-TEST have no monetary value.

Alpha prompts are visible to the contributor and cohort verifiers. Do not submit personal, confidential, sexual, or illegal content.

WORKLOAD256² · 4 steps

No-value alpha credits are escrowed until verifier quorum accepts an output.

MY REQUESTS
  1. No requests from this identity yet.
CONTRIBUTE

Run the next request

Canonical render required

Your GPU is a candidate. Paid claims unlock after the real model pass.

Worker idle

Claiming starts a local render, keeps the lease alive, and uploads the PNG for verification.

OPEN JOBS 0
  1. No queued work yet.
COORDINATORDurable edge ledger
VERIFICATION2-of-N output quorum
SETTLEMENTRobinhood testnet contract gate

DESIGNED FOR TRUST

Every state should explain itself.

01

Local first

Model execution stays on the contributor's machine. Only the result and minimized evidence enter the network.

02

Proof before reward

Canonical qualification and independent output votes make useful work—not time online—the unit that matters.

03

Readable settlement

Queued, claimed, refunded, withdrawable, and settled are distinct states. The interface should never make users guess.