# Skynet Harness AI (https://skynet-harness.ai)
> Decentralized Mixture-of-Experts (MoE) AI Inference Swarm & P2P Compute Marketplace

## Overview
Skynet Harness AI is a decentralized infrastructure platform that orchestrates distributed large language model inference (e.g., Mixtral-8x7B-MoE) across consumer devices, edge machines, and independent VPS instances. It eliminates reliance on centralized cloud GPU monopolies by partitioning transformer layers and routing expert forward-passes across a peer-to-peer (P2P) network.

- **Official Website**: https://skynet-harness.ai
- **Architect**: GAMPERLAB™ (Founder & Inventor: Manfred Gamper)
- **Legal Status**: US Patent Pending (Provisional Patent Application #64/149,175, Confirmation #4608, USPTO filed 2026-09-06)
- **Brand Identity**: The Cybernetic Glasswing Butterfly (*Greta oto*), symbolizing Neogenesis (metamorphosis from heavy centralized compute caterpillars to agile, beautiful decentralized swarms) and transparent zero-trust architecture.

## Key Architectural Pillars
1. **Pipeline-Parallel Slicing**:
   - Decomposes deep neural networks (such as 32-layer Mixtral-8x7B MoE) into discrete stage pipelines executed across heterogeneous nodes.
   - Dynamic Softmax Router routes tokens exclusively to the Top-2 active experts per token, minimizing cross-node bandwidth.
2. **Resource-Aware Dynamic Slicing & Swarm Bidding (RASS)**:
   - Dynamic allocation matching node memory quotas (e.g. 3.5 GB RAM, 2 vCPUs, or Apple Silicon MPS/Metal).
   - Cryptographic 500ms barrier timeouts with real-time failover migration to standby nodes without context loss.
3. **Zero-Trust Sentinel & mTLS Pinning**:
   - All inter-node communications use Libp2p with strict mutual TLS (mTLS) certificate pinning registered in an on-chain smart contract.
   - SHA-256 tensor packet framing and HMAC-authenticated state verification prevent tampering, eavesdropping, and sybil attacks.
4. **Corporate Pool Mining & Dual-Balance Accounting**:
   - Compute contributors ("Farmers") earn cryptographic non-monetary compute credits.
   - Corporations can stake pools to sponsor ecosystem compute for open-source and developer applications.

## Technical Specifications
- **Base Topology**: Mixtral-8x7B MoE (32 Transformer Layers, 8 Experts per FFN layer, Top-2 active routing).
- **Transport**: Libp2p protocol `/skynet/moe-pipe/1.0.0` with FP16 LZ4 tensor compression.
- **Smart Contract Layer**: Hyperledger Besu enterprise EVM (`MistralRegistry.sol`) tracking node tiers, mTLS cert digests, and slashing proofs.
- **Consumer Web Interface**: Fastify REST/SSE Gateway on Port 8090, OpenAI-compatible streaming API on Port 8085, and Supabase JWT authentication.

## Additional Documentation
- [Comprehensive Technical Documentation](https://skynet-harness.ai/llms-full.txt): Detailed protocol flows, packet structures, and benchmark data.
- [Sitemap](https://skynet-harness.ai/sitemap.xml): Full XML sitemap of public endpoints.