Project

ChattyPatty

Decentralized compute LLM inference In the lab Updated

Project by .

Visit ChattyPatty Projects

Decentralized compute marketplace for LLM inference with a Sonic-first trust plane. The architecture pairs an OpenAI-compatible gateway, a scheduler for lease assignment and receipt verification, and provider nodes publishing heartbeat and attestation evidence.

Three services, separate responsibilities
Gateway
Normalize and meter
Scheduler
Assign leases and verify receipts
Provider
Run inference and publish attestations

Receipt commitments bind request, result, and metering. Sonic contracts govern escrow, challenge, and settlement.

On this page

Use

Engineering frame

Constraint
Independent providers cannot be authoritative for their own capacity, token count, or completion of a paid job.
Boundary
The gateway normalizes and meters; the scheduler selects providers and leases; the runtime executes; Sonic contracts govern escrow, challenge, and settlement.
Verification
Canonical commitments bind request, response, model, runtime, tokenizer, and metering. A job cannot complete without a previously committed receipt.

Programming Notes

Current State

In the lab: protocol and specs are locked, contract interfaces and canonical schemas are in place, and gateway/scheduler/provider services are scaffolded and testable.