AI Network and MeshLLM¶
Status
Implemented and tested locally by one operator. There is no public provider network, no public model gateway, and the multi-operator gates are open. Treat everything here as experimental.
| Component | What it is | Status |
|---|---|---|
| MeshLLM runtime | Upstream inference runtime run by providers, pinned and supervised | Local tests pass; distributed routing needs an independent cohort |
| ServiceEscrow | Contract for quotes, prepayment, usage, acceptance, disputes, refunds | ✅ Tested on chain 77240 |
| Revenue policy v3 | How a settled charge is split | Contract-enforced; schedule pending economic approval |
| Wallet AI Sharing page | Opt-in local models, provider policies and P2P overlay | 0.7.3 development branch; complete acceptance not rerun here |
| Wallet Ask flow | Signed quote, explicit funding confirmation, response and escrow acceptance | Implemented in 0.7.3 development; requires eligible providers and testnet funds |
| Training, grants, training rewards | — | Out of scope for testnet-rc1 |
Principles¶
- Never in consensus. No LLM inference is part of block validation. A stopped or compromised runtime cannot stop nodes from validating or moving PQAI.
- Customer-funded. Providers are paid from escrowed customer payments. AI work cannot mint block rewards.
- Off-chain content. Prompts, outputs and weights stay off chain; only hashes, quotes and amounts go on chain.
- Honest receipts. A signed usage statement authenticates the provider's report — not that the work happened.
Developer journeys: AI provider quickstart · Build an AI payment application.
Sources at 72057fe3ec: ai/README.md · contracts/ai/ServiceEscrow.sol · economics/policy-decision-2026-11-05-v3.json · developers/data/feature-checkpoint.json
Found a problem? Report a documentation issue · Build testnet-rc1+72057fe3ec