Whitepaper

Technical specification of the Cynthetic Intelligence protocol.

Abstract

Cynthetic Intelligence presents a decentralized protocol for AI inference that connects GPU providers with AI developers through an OpenAI-compatible API. By leveraging idle consumer and enterprise GPUs worldwide, Cynthetic delivers open-source LLM inference at 5-10x lower cost than centralized alternatives while maintaining quality guarantees through a reputation-based node scoring system and economic incentives powered by the AIC token.

1. Introduction

The AI inference market is growing rapidly, projected to exceed $50B by 2026. However, access remains concentrated among a few large providers (OpenAI, Anthropic, Google), creating bottlenecks in cost, availability, and data privacy. Meanwhile, millions of consumer and enterprise GPUs worldwide sit idle for 80%+ of their operational hours.

Cynthetic Intelligence bridges this gap by creating a two-sided marketplace: GPU owners monetize idle compute as node operators, while AI developers access inference at dramatically lower costs.

2. Architecture

2.1 Overview

The protocol consists of four layers:

  • Client Layer — OpenAI-compatible HTTP API for developers
  • Orchestration Layer — Task routing, node scheduling, billing (Fastify + MongoDB + Redis)
  • Compute Layer — Distributed GPU nodes running Ollama (Electron desktop agent)
  • Settlement Layer — On-chain smart contracts for staking, rewards, and slashing (ERC-20 on EVM)

2.2 Inference Flow

  1. Client sends request to /chat/completions with API key
  2. Orchestrator pre-debits AIC from client balance
  3. Orchestrator selects optimal node by reputation score × stake weight × load
  4. Task dispatched to node via persistent WebSocket connection
  5. Node runs inference locally via Ollama (supports streaming)
  6. Result returned to client; billing settled (refund overpayment)
  7. Operator reward distributed on-chain via RewardDistributor contract

3. Token Economics

AIC (AI Chain Token) is an ERC-20 utility token with a fixed supply of 100,000,000 units. It serves as the medium of exchange for inference payments, operator rewards, staking collateral, and future governance.

The platform applies an 80/20 revenue split: 80% of inference fees go to the processing operator, 20% accrues to the platform treasury for development, liquidity, and ecosystem growth.

4. Staking & Slashing

Node operators may stake AIC via the NodeStaking contract to gain priority in task dispatch. Minimum stake is 100 AIC with a 7-day unstaking cooldown. Operators who consistently fail tasks (3+ failures within 24 hours) are subject to slashing, where a portion of their stake is transferred to a slash treasury.

5. Quality Assurance

Node quality is maintained through a multi-factor reputation system:

  • Reputation score (0-100) based on historical performance
  • Success rate tracking across all completed tasks
  • Load-aware scheduling to prevent overloaded nodes
  • Stake-weighted priority to align economic incentives
  • Automatic slashing for persistent failures

6. Roadmap

  • Phase 1 (Complete) — MVP: inference pipeline, Electron node agent, smart contracts
  • Phase 2 (Current) — Streaming, SDKs, documentation, referral & growth mechanics
  • Phase 3 — Subscription tiers, enterprise SLA, mainnet token launch
  • Phase 4 — Governance, data marketplace, multi-chain deployment

This is a living document. Full technical specification with formal proofs and economic modeling will be published prior to mainnet launch. Last updated: May 2026.