Open frameworks, applications, and tools.
Distributed training framework
A Decoupled DiLoCo–style protocol for verifiable training across heterogeneous GPU contributors. Signed receipts per outer gradient; on-chain run-root commitments per run.
- ·Stake-bonded contributors
- ·Byzantine-robust aggregation (roadmap)
- ·TEE-resident data tier (roadmap)
Inference framework
Reference implementations for tokenized AI inference — chat, vision, audio, forecasting, embeddings, segmentation, detection — with micropayment-channel billing and proof-anchored receipts.
- ·Per-token / per-inference billing
- ·Confidential variants via TEE enclaves
- ·Plonky3 STARK proofs for verifiable outputs
Confidential compute SDK
A unified SDK for running AI workloads inside hardware enclaves with on-chain attestation. Quote verification via the TEE_VERIFY precompile.
- ·Hardware-attested model serving
- ·Encrypted training data residency
- ·Cross-enclave key management
Model marketplace
A first-party application for publishing models to the Tenzro Network — set prices, route inference, accept TNZO, and serve users or agents through the same micropayment rails.
- ·CLI, SDK, and MCP interfaces
- ·Routing across multiple providers
- ·Reputation-weighted selection
Training coordinator
Coordinates verifiable training runs — sponsors fund from on-chain escrow, contributors register stake, aggregators submit signed gradients, and the run finalizes with a public run-root.
- ·Public run roster and receipts
- ·Stake slashing for misbehavior
- ·Cross-region scheduling (roadmap)
Agent runtime
A runtime for autonomous agents on Tenzro — self-sovereign TDIP identity, FROST-Ed25519 threshold wallets, scoped delegations, and policy-bound mandate execution.
- ·AP2 mandates, x402 micropayments
- ·ERC-8004 trustless-agent registries
- ·ERC-4337 v0.8 smart accounts
Agent identity & wallets
Identity and threshold-signature wallets for autonomous agents — issuance, rotation, recovery, and delegation. Built on open standards and registered upstream (CAIP-2, SLIP-44, W3C DID).
- ·did:tenzro method
- ·Hybrid post-quantum signing
- ·Cross-chain identity via TNZO pointers
Mandate & policy engine
Tooling for issuing, scoping, and enforcing agent mandates — what an agent may spend, on whose behalf, for which purpose — with on-chain auditability.
- ·Spend caps and rate limits
- ·Purpose-bound delegations
- ·Revocation and replay protection
Settlement & micropayments
Micropayment channels for high-frequency per-token billing, plus cross-VM settlement across EVM, SVM, and Canton/DAML — one balance, three views.
- ·wTNZO on EVM, SPL adapter on SVM
- ·CIP-56 holding on Canton
- ·LayerZero V2, CCIP, deBridge bridges
Embedded AI cluster runtime
Runtime and scheduler for AI clusters built on embedded AI hardware — the NVIDIA Jetson family and others — supporting model serving, batching, and distributed inference across diverse environments.
- ·Edge, on-prem, mobile, and field deployments
- ·Cluster autodiscovery and sharding
- ·Federated update and rollback
Network bridge
Software that connects a private cluster to the Tenzro Network — registering models, accepting payments, contributing to training, or operating fully air-gapped with selective publication.
- ·Optional public-network registration
- ·Air-gap-friendly modes
- ·Telemetry export, no required upstream
Distributed inference & training
Distributed inference across heterogeneous nodes — pipeline parallelism, model sharding, and streamed token delivery — with the same protocol used for verifiable training contributions.
- ·High-performance inference across clusters
- ·Training contribution from idle capacity
- ·Backpressure-aware scheduling