LIVE on Arbitrum One — real inference settling on-chain. $QAIS is not yet traded on any exchange; acquisition is by direct arrangement only. Your prompts run on strangers' machines; don't send anything secret.
LIVE · ARBITRUM ONEGATEWAY LIVE

Decentralized AI INFERENCE

An OpenAI-compatible API served by independent GPU nodes that earn $QAIS. Every job settles on-chain — 95% to the node, 5% protocol fee.

0
Active nodes
0
Jobs settled
0
Tokens served
0
$QAIS burned
OpenAI-compatible APIArbitrum One mainnet95/5 on-chain settlementEIP-712 batched sessions20% of fees burnedOllama-powered nodesStaked reputation
01

ONE PROMPT GOES IN.

02

INDEPENDENT GPUs COMPETE FOR IT.

03

SETTLEMENT LANDS ON-CHAIN.

lifecycle

How a job flows

Five stages from prompt to payment — coordinated by the gateway, settled on-chain.

  • 01Prompt

    OpenAI-format call hits the API

  • 02Gateway

    matches a staked node

  • 03GPU node

    streams tokens live

  • 04Verify

    oracle samples outputs

  • 05Settle

    95/5 split, one batch tx

See the full architecture →

get started

Two ways in

Use the AI

Point any OpenAI client at the gateway and call it — streaming, a model picker, and pre-funded credit with a single signature.

For developers →

Run a node, earn $QAIS

Got a GPU and Ollama? Install a prebuilt release in ~5 minutes, serve jobs, and earn for every token you stream.

Run a node →

proof of work

Jobs settling on-chain

Every completed job releases escrow in a single batch transaction.

SETTLEMENT FEED
LIVE
  • 0x93f921eb…llama3.213:19:01settled ✓
  • 0x99cccbd8…llama3.213:18:39settled ✓
  • 0xd6a3ebb3…llama3.213:17:21settled ✓
  • 0xac67d464…llama3.220:16:47pending
  • 0x75ca6125…llama3.220:16:45pending
  • 0x09f4584b…llama3.207:55:33settled ✓
why it's real

A real protocol, not a demo

On-chain settlement

Deposit once, sign one EIP-712 spending cap, then fire thousands of calls — they settle in a single batch transaction with zero per-call wallet txs.

Staking + slashing

Nodes stake $QAIS. A 5-dimension reputation score plus Layer-A semantic sampling catch cheaters; disputes slash stake.

How verification works →

Deflationary token

The 5% fee splits 60% ops / 20% stakers / 20% burned. Fixed 1B supply, no mint after launch.

Tokenomics →

Prompt privacy

Verification stores hashes + scores only — prompt/output text is never persisted. ~5% of jobs are re-run on oracle infra to catch cheating.

Trust model today: a single trusted gateway does matching + settlement (worst case is bounded — it can only settle at signed prices, never steal deposits). Removing it — P2P mesh, on-chain auction, decentralized oracle — is the roadmap.

compare

How we compare

Purpose-built for LLM inference — not general compute or rendering, and the only one your OpenAI code can call unchanged.

DimensionQueraISAkashBittensorio.netRender
Primary focusLLM inferenceGeneral cloud computeAI via subnetsGPU clusters for MLGPU rendering + AI
OpenAI-compatible APIYes — drop-in————
Permissionless GPU nodesYesYesYesYesYes
Per-job quality verificationReputation + slashing—Subnet incentives——
Pay with$QAIS / walletAKTTAOWallet / fiatRENDER

Comparison reflects QueraIS's reading of each project's public positioning; these networks evolve fast — check their own docs for specifics.

explore

See the live network

Browse active nodes, watch jobs settle, and chat with the live network — no install, no key needed to look around.

DOCSGITHUB