Conserius
AI disclosure
Eyva — AI Disclosure
Written against the EU AI Act, Regulation (EU) 2024/1689, Article 50, whose transparency duties apply from 2 August 2026.
Eyva is an AI system
Eyva is an artificial-intelligence system. When you talk to Eyva, you are talking to software, not a person. Eyva tells you this itself the first time you open it, before you can use it — the notice cannot be skipped — and in its first words to you:
"Eyva is an AI system. You are talking to software, not a person, and anything Eyva writes for you to send to someone else will say so too."
Eyva may speak in a warm, personal way. That is a manner, not a claim to be a person.
Where its answers come from
- On your Mac: a local AI model (Gemma), which runs entirely on your Mac.
- In the cloud, only if you turn cloud replies on and only where they are offered: a model from Anthropic (Claude), reached through our relay. Which one answered is recorded on your Mac for every turn.
AI output can be wrong. Check anything important. Eyva is not a doctor, lawyer, financial adviser or other professional.
Eyva's voice is synthetic
Every voice Eyva speaks with is generated by a computer. It is not a recording of a person speaking to you. When you turn hands-free voice mode on, Eyva says so in the sentence that confirms the mode.
| Voice | Where it is made | Machine-readable "AI-generated" mark | What the window shows while it speaks |
|---|---|---|---|
| Eyva's own voice (Kokoro) | On your Mac | Yes — the audio carries a mark identifying it as AI-generated | An "AI voice · marked" label |
| The macOS system voice (used when Eyva's own voice is not installed) | On your Mac, by macOS | No — macOS speaks it directly and Eyva cannot add the mark to its audio | A persistent "AI voice" label, which also tells screen readers that the voice is synthetic and carries no mark |
| Premium voice (Inworld) | In the cloud, by Inworld | Yes — its audio passes through the same mark before it plays | An "AI voice · marked" label |
The label appears in the voice and call surfaces of the window the moment a voice starts speaking, and stays there; it names the voice that spoke last.
About the premium voice. It is not active in version 1.0. When it is offered, the text of a reply will be sent to Inworld, one sentence at a time, to turn it into speech. Only replies that already came from the cloud, contained nothing masked and nothing sensitive, will be eligible, up to a daily limit; every other reply is spoken on your Mac. It will be off until you turn it on, and you will be able to turn it off. It will be offered only after Inworld has confirmed to us in writing how it uses and retains that text. See the Privacy Policy, sections 6 and 7.
If you share what Eyva made
If you publish text, audio or images Eyva generated in a way that could mislead people into thinking a person made them — especially a realistic voice or a "deepfake" — the law may require you to say it was AI-generated (AI Act Art. 50(4)). Do not use Eyva's voice to impersonate a real person.
Contact
Questions about this disclosure: support@conserius.com.
© 2026 Conserius Inc. Eyva™ is a trademark of Conserius Inc.