3. Architecture Overview

Osirium follows a modular architecture designed to bridge off-chain AI inference with on-chain validation. The system is composed of four key layers that interact in a seamless pipeline:

🧩 1. Prompt Layer

The user submits a prompt, either manually (via CLI/API) or programmatically (via external sources like X).

🧠 2. Inference Layer

The prompt is processed through an external LLM — currently Claude AI — which generates the response. The AI model is accessed off-chain via secure API and runs in a stateless, verifiable environment.

🔐 3. Attestation Layer

Once the output is received, it is:

• Structured using the EIP-712 standard

• Digitally signed by a registered MCP node (using its private EVM key)

• Returned as a full attestation object (prompt, output, signature, signer)

⚖️ 4. Smart Contract Layer

The signed attestation can be sent on-chain and verified using:

• Verifier.sol – validates the EIP-712 signature

• MCPRegistry.sol – checks if the signer is a trusted MCP node Upon verification, the output becomes an on-chain trusted result usable by other contracts.

This architecture allows Osirium to serve as a decentralized gateway for injecting authenticated AI intelligence into smart contract systems.