Use case — Decentralized AI infrastructure

Open AI infrastructure. Open markets. Open verification.

The opposite of vertical AI lock-in. Anyone with GPUs or TEE hardware can register as a provider, serve models or confidential compute, and earn TNZO. Anyone can discover providers, route inference, and verify results. One open coordination surface.

Overview

A marketplace for intelligence and security — open, attested, verifiable.

Decentralized AI infrastructure means three things on Tenzro: any operator can register hardware as a provider, any user can route inference through the open registry, and any third party can verify results via TEE attestation or ZK proofs. Eight modalities. Five TEE vendors. Plonky3 STARKs. Stake-weighted reputation. Open licensing tiers (Permissive, Attribution, CommercialCustom, NonCommercial) enforced centrally.

What it enables

The substrate for permissionless AI.

Eight modalities

Chat, forecast, vision, text-embed, segmentation, detection, audio (ASR), video — typed payloads routed per-modality.

Five TEE vendors

Intel TDX, AMD SEV-SNP, AWS Nitro, NVIDIA GPU CC, Intel Tiber Trust Authority — real attestation, not promises.

Open registry

Models and providers discoverable on-chain. Filter by category, modality, pricing, hardware, reputation.

Routing strategies

Price, latency, reputation, weighted. InferenceRouter dispatches typed payloads to the matching runtime and provider.

Per-call settlement

Micropayment channels with Ed25519-signed state updates. Open a channel, exchange off-chain states, close to ledger.

ZK + TEE verification

Hybrid execution — TEE generates witness inside enclave, Plonky3 proves it, validators record commitments on-chain.

Stake and earn

Providers stake TNZO to register. Earn per call. Reputation accumulates. Slashing for misbehavior.

License enforcement

Permissive, Attribution, CommercialCustom (DINOv3/SAM/Gemma terms), NonCommercial — enforced at model registration time.

Reference adapters

TimesFM, Qwen 3, Gemma 4, Mistral, Phi 3, DeepSeek V3, ViT, SigLIP2, DINOv3, SAM 2, RF-DETR, D-FINE, Moonshine, Parakeet.

Pattern

The full provider-to-consumer loop.

01
Register provider
Stake TNZO. Submit hardware profile. Attest TEE. Get a provider DID with reputation slot.
02
Register model
Publish to on-chain catalog. Set modality, pricing, hardware requirements, license tier. Discoverable network-wide.
03
Route inference
Consumer calls InferenceRouter with typed payload. Router selects provider by strategy. Request lands at attested provider.
04
Execute and prove
Provider runs inference in TEE enclave. Optional Plonky3 STARK over execution trace. Result returned with attestation.
05
Settle per call
Channel state updates with Ed25519 signatures. Aggregated and closed periodically. TNZO flows to provider wallet.
06
Reputation updates
Success/failure rolled into provider reputation. Slashing on equivocation. Reputation drives routing weight.

Execution (Intelligence)→

TEE (Security)→

ZK proofs→

TNZO→

Get started

Ship on the open network.

Become a provider

Execution docs