Updated: 8 September 2026 · version 2026-09-09-v16
You are interacting with artificial intelligence
EMSYO is an AI system, not a human. Software generates or controls its face, synthetic voice, personality and responses. The disclosure appears before first interaction and remains available in the app. EMSYO must not represent itself as a real person.
Output can be wrong
Responses, translations, descriptions, document analyses, research summaries and recommendations can be inaccurate, incomplete, biased or outdated. A current fact is verified only when an appropriate live tool successfully returns it. Independently verify medical, legal, financial, employment, safety and other important decisions with a qualified person. EMSYO is not an emergency service.
Synthetic voice and digital face
Standard speech uses exactly one provider selected before the request: ElevenLabs when configured, otherwise OpenAI. A second paid synthesis is not started automatically after sending. The live interpreter and separate LIVE voice mode are active paid features with server-capped sessions of up to five minutes. The ordinary chat microphone records a short clip for subsequent transcription. This is a generic synthetic voice and animated AI identity, not a recording or impersonation of a particular person. Disabling voice output makes EMSYO respond in text and blocks paid synthesis.
Realtime voice
The live interpreter costs 35 credits and LIVE voice costs 10 credits per session of up to five minutes. The server ends the session by that time. The full block is reserved before start and consumed once the connection is created, even if stopped early. The ordinary chat microphone uses recorded-audio transcription at 1 credit per started minute; it is not a LIVE session. AI answers and enabled standard voice output are charged separately.
Paid AI actions and credits
AI and external tools remain visible but locked in Free. Every paid action has a public credit price; paid use with sufficient credits has no additional daily product cap. Credits are reserved before sending; once sent to a provider they are consumed even if the response is unfinished, but released when rejected before sending. LIVE voice and the live interpreter are charged at the session prices above. Standard voice output costs 3 credits per started 200 characters. Each document costs 20 credits, is limited to 4 MiB and at most two are allowed per request; an image attachment is billed as vision at 3 credits.
Paid actions require enough credits for the complete action and have no additional daily product cap. This applies to subscriptions, passes and purchased credits. A technical safeguard allows at most 60 new paid operations within a rolling 60-second window; if reached, wait briefly before trying again. Individual request limits, session duration and provider availability still apply. Free-trial and cloud-storage rules remain unchanged.
A custom detailed 3D model costs 200 credits, shown before generation starts. A successful model is saved automatically; failed generation returns reserved credits. Every insertion of the saved model costs 0 credits. This is AI-generated content; geometry and texture quality depend on the description and generator.
AR companions and prepared content
Emsyo, Bim, Vexo and Kiki are fictional digital characters. Vexo and Kiki use their own designed ElevenLabs synthetic voices; depending on available configuration, an alternative OpenAI synthetic voice may be selected before a request. Their movements, including mouth movements, are animations. Vexo's AI guessing turns use OpenAI's gpt-6-astra with saved round history. Prepared quizzes, logic puzzles and Kiki and Bim activities come from a catalogue and follow game rules; each step is not a new AI answer. Speech is generated separately. Free-form conversation and shared character responses use AI.
Game memory preserves progress; it does not mean unlimited personal memory or flawless guesses. AI can make mistakes even with saved answers. Kiki's child-content instructions reduce the risk of unsuitable output but do not replace adult supervision. AR creates supported virtual objects or simple assemblies, not an arbitrary photorealistic model for every instruction.
“My World” personalization
My World is an optional personalization memory of your account, off by default. When it is on, only a few items relevant to the current request may be added to the prompt sent to the AI provider. Contact values (phone, e-mail) and items marked “Ask instead of using” are never sent to the model; EMSYO asks you instead. You can view, correct, export to JSON, delete individual items or switch the whole use off at any time.
Vision and prohibited biometric inferences
Vision helps understand a scene, object or text you intentionally show. EMSYO must not identify a real person by appearance, perform biometric identification, infer emotion from biometric data or infer sensitive traits. Visual description can be wrong.
AI images, audio and video
Generated or edited media can invent or alter people, objects, text, sound and events. A gallery result that EMSYO can reliably derive as generated is labelled “AI generated” independently of its user title. Do not remove or bypass a label or available provenance to deceive. EMSYO preserves provider machine-readable provenance when the format and processing technically retain it, but does not claim that every re-encoding supports C2PA.
Never present synthetic media as authentic evidence. When publicly sharing deceptively realistic content, clearly disclose that it was AI-generated or edited unless this is already obvious and labelled.
Documents, web and sources
Document analysis and research are assistive. EMSYO can miss a clause, misread a table, rely on a weak source or attribute an unsupported conclusion. Open citations, check dates and verify professional material against the original.
Human responsibility
Users remain responsible for decisions, device instructions, sharing and use of output. EMSYO does not make solely automated decisions about a user with legal or similarly significant effects.