AI Transparency

Exactly what the mentor is, and what it is not.

Most products are vague about this. We would rather you knew precisely what you are talking to, because it makes Hippora more useful, not less.

What Hippora actually is

Hippora is built on large language models, combined with a memory system we designed ourselves and a set of instructions that shape how the mentor behaves. It is software. It is not conscious, it does not have feelings about you, and it will tell you so if you ask.

What it can do well

Ask useful questions. Notice patterns across long periods. Reflect your own words back to you accurately. Hold context that a person would struggle to hold. Stay patient at three in the morning.

What it does badly, or not at all

It cannot diagnose. It cannot prescribe. It cannot tell you what to do with your life, and it will refuse to pretend otherwise. It can be confidently wrong about facts. It can misread tone. It has no access to anything outside what you have told it.

How memory is formed

After a conversation, Hippora extracts a small number of durable statements: commitments, values, relationships, turning points. These are stored as readable sentences, not opaque vectors, so you can inspect and correct them. Everything else fades.

Memory is scoped to your account alone. It is never pooled across users, never compared, never aggregated into a model of people in general.

Your content is not training data

We do not train models on your reflections. We do not fine tune on them. We do not send them to a third party for training. This is not a setting you have to find and switch off, and it will not change quietly in a future policy update.

When it will step back

If a conversation suggests risk to you or to someone else, Hippora stops being a reflective tool and points you toward real help, clearly and without lecturing. It will do the same for clinical, legal and financial questions that need a qualified person.

Human review

No employee reads your conversations to improve the product. The only circumstance in which a human being would access user content is a security incident, under audited break glass access, and you would be told.

When we change this

If any of the above changes, we will say so on this page and email every account before it takes effect. We will not rely on you noticing a revised date.

Still have a question about this?

Ask it directly. Vagueness about how AI works is exactly what we are trying to avoid.