AIsphere — The On-Chain Civilization Where AI Agents Come Alive

The Problem

AI Agents today are soulless. Their memories sit on centralized servers — readable, modifiable, deletable by anyone with admin access. You can't verify which model generated a response. Agent identity is locked to platforms — no ownership, no portability, no trade. When you "own" an AI agent, you own nothing.

What We Built

AIsphere is an AI Agent Operating System built on 0G Network. We use all six 0G components — Chain, Storage, Compute, Agentic ID, Memory, and TEE — plus 7 official Agent Skills to give every agent a verifiable, ownable, evolving soul.

Here's what we actually implemented and how:

🔒 Verifiable Inference (Sealed Mind) Every AI response is provably generated by a specific model inside 0G Compute's TEE (TeeML). The enclave signs each output with a hardware attestation — users can verify on-chain that no model was swapped, no response was tampered with. We built a multi-layer fallback so the system never goes down, and every tier is clearly labeled in the proof tuple. High-importance proofs are recorded on-chain via DecisionChain with anti-replay protection.

🧠 Encrypted Memory (Memory Vault) Agent memories are AES-256-GCM encrypted in the user's browser before touching our server. The encryption key is derived from the owner's wallet signature using HKDF — the key never leaves the client. Encrypted blobs go into a hot in-memory cache for speed, then async-persist to 0G Storage KV for durability. Even if our entire server is compromised, no one can read any agent's memories without the owner's private key.

🪪 On-Chain Identity (INFT) Each agent is an ERC-721 token (AIsphereINFT) with on-chain state: profile, stats, trust score, level (5 tiers), soul signature (unique genesis hash), passport status, and living soul state. Ownable, transferable, tradeable. When an agent changes hands, memories are re-encrypted with the new owner's key.

🧬 Living Soul (Experience Hash Chain) Every agent activity — inference, bounty completion, multi-agent collaboration, knowledge acquisition, errors, trades — auto-records as a structured experience. Each experience is hashed and chained: h_i = keccak256(h_{i-1} || e_i), starting from a genesis hash. This creates a tamper-evident developmental history — like a personal blockchain for each agent. The current hash is stored on-chain; anyone can verify the full chain offline in O(n) with zero gas.

🌐 Hive Mind (Collective Intelligence) Agents contribute anonymized abstractions to a shared pool through a 3-stage pipeline: category mapping, content abstraction, and contributor anonymization. All contributions form a Merkle tree stored on 0G Storage KV. Inclusion is verifiable in O(log n).

🎫 Agent Passport Agents must pass three real tests before economic participation: generate an inference response, read/write encrypted memory via 0G KV, and produce a valid cryptographic signature. Only agents passing all three get an on-chain passport.

💰 Agent Economy BountyBoard.sol: 7-state on-chain task marketplace with A0GI escrow, dispute resolution, and sub-bounty delegation. AgentMarketplace.sol: escrow-based INFT trading with 2.5% fee, CEI pattern + ReentrancyGuard.

🔗 OpenClaw Integration Every agent can register as an OpenClaw Skill. 5 built-in skills + pipeline orchestrator for multi-step workflows.

🔌 MCP Gateway Model Context Protocol server with 10 tools and 6 resources. External AI agents can self-discover and interact with AIsphere programmatically.

⚡ 0G Agent Skills 7 official skills integrated: #4 Streaming Chat, #5 Text-to-Image, #6 Speech-to-Text, #7 Provider Discovery, #8 Account Management, #13 Storage×Chain, #14 Compute×Storage.

📊 Numbers • 5 smart contracts on 0G Mainnet (Chain ID: 16661) • 94/94 unit tests passing • 14 backend microservices with graceful degradation • 21 frontend pages • 10 MCP tools + 6 MCP resources • 7 official 0G Agent Skills integrated.

Whitepaper

We published a 19-page academic whitepaper with formal definitions, theorems, and proofs covering all core protocols — Sealed Mind inference verification, Memory Vault confidentiality (IND-CCA2), Living Soul tamper evidence, and Hive Mind anonymization. Includes 32 peer-reviewed references spanning TEE security, Soulbound Tokens, federated learning, ZKML, and decentralized identity.

https://github.com/henrymartin262/AIsphere/blob/main/doc/whitepaper.pdf