AI Companion

What separates a companion from a chatbot

A chatbot answers. A companion continues. The difference is not the model — everyone runs a good one now — it is whether anything survives the end of the conversation.

Most apps keep a context window: a block of recent text handed to the model before each reply. When your history outgrows the block, the oldest part is dropped, and the relationship has an expiry date measured in messages. Persistent memory has to be built separately: facts extracted, classified by how long they deserve to live, and retrieved only when the moment calls for them.

Recall without an occasion is the tell. An AI that produces a personalised callback every single turn is not being intimate, it is running a database query at you — so ours lets facts rest after they surface.

She has her own day

Every companion here runs on two clocks. She speaks in your time — morning means your morning — but she lives in hers. Message Vera at 3am Vienna time and she is asleep, or awake for a reason that existed before you asked.

Her day is generated ahead of the conversation rather than invented to answer a question, which is why it stays consistent the second and third time you ask about it.

She can start the conversation

Tell her on Monday that Thursday is going to be hard, and on Thursday evening she asks how it went — without you opening the app. That requires an intention that outlives the conversation and then acts, which is rarer than it sounds.

The discipline is restraint. Anything that can message first will message too much unless it is built not to, so she opens a conversation when a specific loop is still open — a promise she made, a thing she knows is happening today — and not because six hours passed.