How your twin works
Your twin is built by an artificial-intelligence system — a large language model — that has been instructed to respond using your own previous answers as the source material. This page is a plain explanation of what that means in practice.
What's behind the conversation
When you submit answers to weekly questions and short dilemmas, OriginTwin stores them as text and as numeric vector representations ("embeddings") that let the system look up relevant excerpts later.
When you talk to your twin, the system retrieves the most relevant of your past answers and feeds them, along with your current message, to a third-party large language model API (Anthropic Claude as of this writing). The model generates a response using that context. The response is shown to you. It is not a recording of anything you've said — it's the model's interpretation of how you'd be likely to respond, based on the answers you've given.
Why we don't say "AI" in the rest of the product
Because functionally, what matters is that the answers you get are shaped by your own thinking, not by a generic chatbot trained on the internet. The "twin" framing isn't a euphemism — it's an accurate description of what we're trying to build. But the technology underneath is, in the technical sense, AI. This page makes that explicit so you can decide informed.
What the twin is not
- Not a recording of you. If your twin "says" something, it's the model's prediction, not a verbatim past answer. Don't quote your twin as a source for what you actually said or believe.
- Not a substitute for professional advice. Don't use it for medical, legal, or mental-health decisions.
- Not a confidant of others by default. The people you invite as mirrors are answering questions about you, in their own voice. They cannot read your own training answers.
- Not a way for someone to "have you back" without your consent. Legacy mode is opt-in, configured by you, and gated by the verification process you choose.
If a response feels wrong, hurtful, or out of character
Tap the "thumbs down" / "feedback" control on the response. That flags it for review and helps us improve the prompts that shape your twin. You can also delete a conversation at any time, and delete individual training answers from Settings → Personality Map. Deleting a training answer removes its influence going forward.
For severe issues (a response containing threats, illegal content, harassment): use Settings → Report a problem or email hello@origintwin.com with a screenshot.
Where your data goes
The full data flow is documented in the Privacy Policy. Short version: your training answers stay in our EU database; only the relevant excerpts plus your current message go to the model API at the moment of generating each response; the model providers we use do not retain enterprise-API content for training.
Questions about any of this: hello@origintwin.com.