The Siggy, explained

Ritual is a chain purpose-built for AI infrastructure. Instead of contracts calling out to centralized AI APIs, Ritual runs inference natively at the protocol level through 16 precompiled contracts, executed inside TEEs and cryptographically bound to the exact request that triggered them.

The Siggy calls the LLM Precompile with a locked system prompt that turns it into a metadata engine: it only ever returns structured JSON trait data (fur, eyes, hat, glasses, clothing, accessory, expression) chosen from an allowed list. It never writes image prompts or describes art — that boundary is what keeps every Siggy visually consistent.

Once traits come back, the frontend builds a single locked pixel-art prompt — same camera angle, same lighting, same Ritual-green background — and sends it to the Image Precompile. This is a long-running async call: the job is tracked via Ritual's AsyncJobTracker and delivered back through AsyncDelivery once rendering completes.

No OpenAI, Claude, Stability, or HuggingFace endpoint is ever called. Both the metadata engine and the image renderer run as on-chain precompiles on Ritual Testnet, so the entire creative pipeline — decision-making and rendering — is verifiable and reproducible on-chain.