AI Disclosure
Last updated: 28 Oct 2026
Voxiora is an artificial intelligence system that conducts phone and web conversations on behalf of businesses. Everyone has the right to know when they are talking to AI; this page explains how we ensure that and how the assistant works.
1. The assistant always says it is AI
Every conversation conducted by Voxiora starts by telling the caller that they are speaking with an AI assistant, in accordance with Article 50 of Regulation (EU) 2024/1689 (the Artificial Intelligence Act). Businesses can customise the greeting, but a greeting without this notice cannot be saved, and the assistant is prohibited from claiming to be human.
2. What the assistant does
It understands speech, answers by voice based on the business’s knowledge and instructions, books appointments, takes messages, sends confirmations and transfers calls to people. It does not make decisions that produce legal effects concerning callers or similarly significantly affect them.
3. Recordings
The business decides whether calls are recorded. When recording is enabled, the greeting tells the caller. The business also decides how long recordings and transcripts are kept, from deletion the day after the call to storage until deleted manually.
4. Limitations
AI systems can misunderstand what is said or make mistakes. Voxiora is instructed to answer only on the basis of the business’s knowledge, to admit when it does not know something and to hand the call over to a person when the caller asks or the matter requires it. The assistant must not be used to handle emergency numbers.
5. Speaking to a person
The caller can ask to speak to a person at any time. Depending on the business’s settings, the call is transferred or the assistant takes a request for contact.
6. Technology and data
Voxiora combines speech recognition, large language models and speech synthesis. Speech recognition is provided by Soniox (with a model running on our own servers as a backup), responses are generated by the Google Gemini (via OpenRouter) and Anthropic Claude language models, and the assistant’s voice is synthesised on our own servers in Poland. Call data is not used to train AI models. The full list of providers is on the subprocessors page.
7. Assistant voices
Catalog voices are built from recordings by people who agreed to synthetic voices being created from them, published under open licences: Kyutai Unmute Voice Donation (CC0), Open Home Foundation voice datasets (CC0) and the CSTR VCTK Corpus 0.92 — Junichi Yamagishi, Christophe Veaux, Kirsten MacDonald, University of Edinburgh, Centre for Speech Technology Research, licensed under CC BY 4.0. Additionally: the Alba speech corpus (CSTR, University of Edinburgh) and the SIWIS French Speech Synthesis Database (Idiap Research Institute), both licensed under CC BY 4.0. The Kamil voice: The MC Speech Dataset (Mateusz Czyżnikiewicz, Warsaw University of Technology 2022, OpenSLR 142, public domain — a speaker's recordings made for Polish speech-synthesis research). The voices are synthesised — they are not recordings of these people and do not speak on their behalf.
8. Contact
Questions about how the AI assistant works can be sent to support@aiagenthub.pl.
AI AGENT HUB PROSTA SPÓŁKA AKCYJNA · support@aiagenthub.pl
ul. Jana III Sobieskiego 11/E6, 40-082 Katowice · NIP: 6343054493 · KRS: 0001168165 · REGON: 541503506