PQWERAI

AI Compute

Inference you pay for only when it is delivered.

Providers run MeshLLM outside consensus, quote a price, and are paid through ServiceEscrow. Inference never enters block validation.

CONNECTINGLast update

Share AI power. Get paid for accepted work.

A client locks PQAI in escrow, a provider runs the model on its own hardware, and payment is released only when the client accepts the result; otherwise the client is refunded. Models never run inside block validation.

AI market: Deployed on testnetModel pools: Deployed on testnetMeshLLM/Colibri: Tested locally

Providers (on-chain registry)

Registered on chain
CONNECTING
Enabled
CONNECTING
Reachable from this site
CONNECTING
Registry
CONNECTING
First provider
CONNECTING
Its model
CONNECTING
Queue, inference speed, uptime
UNAVAILABLE — not reported by the provider endpoint

Models

Approved model files on the read-only gateway (reachability: one-file gateway probe)
FileSizeGateway servingChecked

Model pools

Same-model compute pools (CR-0033)

What a pool is
PCs that run one specific model together and add up their capacity. Pools are created and joined from the wallet's Model Pools page or the AI Sharing page (wallet 0.8.7).
AI market on testnet 20777
LIVE since 15 Nov 2026: ModelRegistry 0x0d732d4d4521341e98bb294855837BEBE23Dc1Fa (the wallet model is registered and active), AIJobMarket escrow 0xd5ACf35F5835e55B8e4328a51720D9F27836C8e4; accepted jobs pay 88% to the provider and 4% each to gateway verification, the CPU pool and the Smartnode pool. Unaudited test-network contracts.
Pool directory
LIVE three federated directories (seed1, seed2 and POOL1) replay the signed pool log; public read API https://pools-testnet.pqwerai.com/v1/pools and pools-testnet2; the explorer has a Pools page
Relay for PCs behind NAT
LIVE iroh-relay on the pool hosts; admits only members that a directory lists as ready or active
Pool power and penalties
LIVE since 18 Nov 2026 (CR-0066): members send heartbeats; the directory shows pool power and records reputation strikes for leaving early or going silent. An on-chain bond comes later.
Pools right now
ONE bonsai-2-community (Bonsai 2 2B, open to GPUs with 2 GiB or more, price 0), created 18 Nov 2026; no outside members yet. Create or join a pool
Splitting one model across PCs
OPEN two-stage split verified on one host/GPU; a run across several independent GPU PCs is not verified yet
Pooled settlement
NOT IMPLEMENTED — pool members are not paid on chain yet
Technical details, live data and component status
Become a provider
Provider onboarding
AI API
AI network · ServiceEscrow · Build an AI payment app
Settlement events
Explorer AI market events
Revenue policy
Revenue policy v3 — contract-version specific; no founder cut on AI revenue
AI components — verbatim from the component registry, updated 2026-11-19
ComponentStatusRegistry ID
MeshLLM runtime adapter and supervisorPARTIAL (G5-REAL-MESH-INFERENCE open)CMP-MESH
AI provider agent (GPU worker, transport, request auth)LOCALLY VERIFIED (NATIVE-AI-* closed local); independent providers openCMP-AI-PROVIDER
AI customer routing, registry client and chain guardsLOCALLY VERIFIED (NATIVE-AI-ROUTING-POLICY-LOCAL)CMP-AI-ROUTING
AI settlement, finality and revenue policyPARTIAL (G6-VERIFIED-SETTLEMENT open)CMP-AI-SETTLEMENT
ServiceEscrow v3 (+ immutable legacy v1/v2)IMPLEMENTED v3 CONFIGURABLE; legacy terms immutableCMP-ESCROW
AI P2P overlay daemon (libp2p, relay, DCUtR, NAT traversal)PARTIAL (AI-P2P-* findings: PCP, real external relay, Windows home provider)CMP-AI-P2P
Cooperative GPU pool managerPARTIAL (CR-0018, CR-0033): signed registry, membership epochs, planner, contribution accounting and per-model pool directory with wallet create/join LOCALLY VERIFIED (control plane); upstream two-stage split of one request executed on one host/GPU; independent multi-host, heterogeneous CPU stage, topology-lock handoff and pooled on-chain settlement openCMP-POOL