the protocol
WAGE is a decentralized AI inference market on Solana. Buyers pay USDC for credits and redeem them for inference jobs. Workers run a lightweight WebGPU SDK in a browser tab and earn USDC for completed jobs. The protocol captures a thin margin and uses it to buy back and burn $WAGE while paying USDC yield to $WAGE stakers. No emission rewards. No fake APY. The work is the work.
economy
RAILS
The buyer pays in USDC. Credits cost $0.01 each. Inference jobs cost between 1 and 12 credits depending on type and model size. The worker is paid in USDC, settled per job (Phase 1: 24h batched, Phase 2: per-claim, Phase 3: per-job onchain).
SPLITS
| recipient | share | token |
|---|---|---|
| worker (the rig) | 70% | USDC |
| treasury | 30% | USDC |
| treasury → $WAGE buyback & burn | 15% | $WAGE burned |
| treasury → $WAGE staker yield | 15% | USDC |
$WAGE TOKEN
| property | value |
|---|---|
| total supply | 1,000,000,000 |
| team allocation | 0% |
| presale | 0% |
| launch venue | pump.fun bonding curve |
| liquidity | locked at TGE, raydium |
| burn rate | per inference revenue, no schedule |
worker SDK
The worker is a lightweight WebGPU client that runs in a browser tab or a headless Node process. It accepts inference jobs from the protocol queue, runs them locally, returns the result, and gets paid. The MVP supports four job classes.
SUPPORTED JOB TYPES (v0.1.2)
| job | model | vram | typical job time | credit cost |
|---|---|---|---|---|
| embed-text | all-MiniLM-L6-v2 (q4) | ~400MB | 0.2 – 0.6s | 1 |
| classify | distilbert-base (q4) | ~600MB | 0.3 – 0.8s | 1 |
| image-tag | clip-vit-base (q8) | ~1.2GB | 0.4 – 1.2s | 2 |
| small-llm | llama-3.2-1b (q4) | ~1.6GB | 1.5 – 4.0s | 4 – 12 |
| ocr-snippet | tesseract.js + crnn | ~250MB | 0.5 – 1.5s | 2 |
TECHNICAL REQUIREMENTS
- browser: Chrome/Edge 113+ or Firefox 121+ (WebGPU required)
- os: Windows, macOS, Linux. Mobile excluded in v0.1 (handled in v0.4)
- minimum: integrated GPU + 4GB shared VRAM (CPU-only fallback at half rate)
- recommended: discrete GPU + 6GB VRAM + 50 Mbps connection
- quantization: GGUF q4_K_M and q8_0 supported via wllama and transformers.js
- job timeout: 30s per job; jobs that exceed timeout return to queue
- connection: WebSocket to nearest router, fallback to long-polling
- reputation: worker keeps a per-job latency + correctness score; bad scores reduce job routing
RUN IT
Open the rigs page. Connect a Solana wallet. The browser detects your GPU. The protocol enrolls you as a worker. The shift begins.
For headless deployment (Linux server, raspberry pi farm, cron worker), the SDK is published as @wage/worker:
npx @wage/worker --wallet WAGEXX...XX --union SECTOR-7-LOCAL-001
onchain phases
PHASE 1 · OFF-CHAIN MVP (now)
Worker daemon runs in browser. Job routing, queue, billing, scoring all off-chain. Token launches on pump.fun. Wallet sign-in only. Payouts batched and settled to wallet every 24 hours. This phase exists to prove the unit economics on a real workload, not to be cypherpunk.
PHASE 2 · CLAIM ONCHAIN (Q3 2042)
Workers sign claim transactions onchain with anti-replay slot-hash binding. Treasury runs the buyback program through a Jupiter-routed program with weekly cadence. $WAGE staker yield distributed onchain per epoch.
PHASE 3 · FULL ONCHAIN ROUTING (2043)
Job queue itself runs as a Solana program with worker registration onchain. Each job is a state transition. Each completed job is a compressed NFT receipt for both buyer and worker, permanent. Latency-sensitive jobs route off-chain with periodic onchain anchoring.
governance
$WAGE stakers vote on protocol parameters (margin split, supported model list, union charter rules) via weekly snapshot. No off-chain DAO. No multisig escape hatch. Code is law except where the work demands otherwise.
fair launch
Token launches on pump.fun. No presale, no whitelist, no allocations to the team beyond what they buy on the open market like any other worker. Liquidity is locked at TGE. The first week of inference revenue funds the first buyback. The first burn is publicly verifiable.
faq
WHY NOT BUILD ON ETHEREUM
Solana's slot rhythm (400ms) matches the inference job cadence. Settlement cost on Solana lets per-job claims be economically viable. The buyer doesn't care which chain; the worker cares about settlement latency.
WHY GAMEIFY IT
Existing decentralized compute networks have a friction problem: install a CLI tool, configure ports, watch a JSON log. Workers churn. WAGE wraps the same compute in a UI that returns a tactile signal for every job. The work doesn't change. The retention does.
WHY NOT JUST USE c0mpute
c0mpute solved the infra. WAGE solves the audience. Same kalup (USDC payment, $TOKEN buyback, no emission), different gravity. Two networks can coexist. The buyers care about latency and price, not branding.
WHAT IF THE NETWORK HAS NO BUYERS DAY 1
Day 1 is bootstrapped with a small set of seeded buyers (initial agentic apps that pay for inference). The protocol absorbs the initial cost as $WAGE-burn-positive activity. The flywheel starts at five real buyers, not zero.