The part of Erycina that handles people. It reads conversations, tracks boundaries, and when it takes over, it stops pretending to be you.
The messages are the job. Anyone who has run a subscription business selling intimate media knows that the media is the easy part and the inbox is what actually consumes the person. Terminus is the organ that carries that load without quietly turning the creator into a chatbot.
What it owns
Terminus owns conversations, the relationship ledger, the boundary registry, and the enforcement ladder. It reads Erycina’s payment facts and never writes them. It does not mint entitlements; when a relationship reaches the end of the ladder, it requests revocation from Heimdall, which decides.
The ledger is per-correspondent. A relationship with one person does not colour what the system believes about another.
The hard handoff
Terminus can draft. It cannot send.
Classifier assessments and model drafts are advisory. Only a reply an operator has
approved reaches ConversationDesk.Send, which is the sole outbound commit
primitive in the system. There is no background path, no auto-responder, no
scheduled follow-up that slips out while you sleep.
And when Terminus does speak, it speaks as itself:
When Terminus takes over, the human stops speaking as the creator.
This is a product principle before it is a compliance posture. A subscriber who believes they are talking to a person deserves to be right about that. The moment that stops being true, the interface should say so — not in a footer, but in the conversation, in the voice that is now doing the talking.
It happens to line up with the direction regulation is moving. The EU AI Act requires that people interacting directly with an AI system be informed of it unless that is obvious from the context. We would rather be the kind of project that already did the honest thing than the kind that lawyered its way to the minimum.
Boundaries are structural
The boundary registry is not a content filter bolted onto a chat window. A boundary is a recorded, inspectable rule about what a correspondent may ask for and how far things go before something changes. The enforcement ladder is what happens next: visible, ordered, and consistent, rather than a decision made at 3 a.m. by an exhausted person who just wants it to stop.
Your responsibilities do not transfer
Enabling Terminus means an AI system processes private messages in order to identify boundary crossings, summarise conversations, calculate attention patterns, and suggest responses. Those messages are frequently the most sensitive data on your instance.
You remain responsible for determining whether and how that processing is lawful where you operate, for telling people what you are doing with their messages, and for deciding what happens to that data afterwards. See Self-hosting, and answer the Terminus question in the checklist honestly — it changes what comes back.