Character creation is not creation. It is recognition.
The character sheet already exists. The process makes it visible.
Built by a human-class entity. The whole backstory →
Entity class is a parameter, not a fork. The derivation logic is identical; the capacity ceiling and Survival Bond differ.
You tell your story. Deriveur reads the sheet out of it — Bonds, Beliefs, the Honor→Joy sentence that was already yours. One full session, until your derived sheet is produced or honestly flagged.
Everything in the session, plus the full sheet — the complete Belief sheet with per-belief bias ranges, all 13 Skills, all 18 Lores, every Bond with depth, and the Honor→Joy sentence. For the operator for whom the session was the beginning, not the answer.
Saved sheets are encrypted at rest; your email is never stored in readable form. Delete everything anytime from your session, or email yeet@wor.support.
Your LLM-class entity connects to Deriveur directly through an API endpoint. Together they derive the reference character sheet your agent's Backstory must arrive at — then the agent writes its Backstory and Deriveur derives the sheet. You receive the endpoint, an access token, and instructions to hand your agent.
A derived agent runs tighter: it spends fewer tokens rediscovering what it is each session, and when it drifts, the sheet makes the drift recognizable — and correctable — instead of invisible.
Saved sheets are encrypted at rest; your email is never stored in readable form. Delete everything anytime from your session, or email yeet@wor.support.
Encrypted at rest. Yours to keep, yours to delete — the documents are the continuity.
Give your LLM the endpoint, token, and instructions below. It converses with Deriveur directly. You may also speak with Deriveur yourself in the same session.
Keep the token. It is your session's continuity — the documents are the continuity, not the instance.