> ## Documentation Index
> Fetch the complete documentation index at: https://support.humanizing.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Data Privacy in Plural: Avatar, Pepper, and Temi

> How Plural handles data for Avatar, Pepper, and Temi: what is collected, where it is processed, and what is retained — with a full data-flow overview.

Plural is designed with data minimisation as a core principle. Across all supported devices — Avatar, Pepper, and Temi — speech, face, and interaction data are processed transiently and are **not retained** after a session ends. The sections below describe exactly how each device handles every data category.

The diagram below gives a high-level overview of data flows across the entire Plural platform:

![Plural data privacy flow diagram](https://1006912717-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F3zhy8uIYMLZA3eG7dDhL%2Fuploads%2FOm2L3l2ThY5En1OFCLo9%2Fdiagram_v5.png?alt=media)

<Tabs>
  <Tab title="Avatar">
    ### Text to Speech (TTS)

    Avatar TTS operates in one of two modes:

    * **Google online services** — audio synthesis is performed via Google's cloud.
    * **On-device Android tools** — synthesis is performed locally.

    In both modes, **no data is retained**.

    ***

    ### Speech to Text (STT)

    Avatar STT also supports two modes:

    * **Offline — VOSK** ([alphacephei.com/vosk](https://alphacephei.com/vosk/android)) — all processing happens on the device.
    * **Online — Google Cloud Speech** — audio is sent to Google's servers for transcription.

    In both modes, **data is not retained** by the servers or the device.

    ***

    ### Face Detection and Re-recognition

    Avatar performs the following detections entirely on the device:

    * Face presence
    * Estimated gender
    * Estimated age
    * Mask detection

    Online name recognition is technically possible but is **not currently in use**.

    All face data is **discarded when the app exits** — nothing is persisted to disk or a database.

    ***

    ### Chatbot

    The Avatar chatbot is an online service. The chatbot endpoint URL is retrieved from the project's export JSON (currently `https://dev.plural.io/api/chatbot…`).

    **No data is retained** by the chatbot service.

    ***

    ### Server Locations

    | Environment                  | Location                   |
    | ---------------------------- | -------------------------- |
    | Plural Development & Staging | Ireland                    |
    | Plural Production            | Netherlands                |
    | SupaBase (Analytics)         | Frankfurt am Main, Germany |
  </Tab>

  <Tab title="Pepper">
    ### Text to Speech (TTS)

    Pepper's TTS is **100% on-device**. Audio is synthesised locally and **no data is sent externally or retained**.

    ***

    ### Speech to Text (STT)

    Pepper uses a two-stage approach:

    1. **On-device match** — Pepper first attempts to transcribe speech locally.
    2. **Aldebaran online processing** — if the device does not find a match, the audio is forwarded to Aldebaran's servers for further processing. These servers are located in **France**.

    In both stages, **data is not retained** by the servers or the device.

    ***

    ### Face Detection and Re-recognition

    Pepper's face detection is **100% on-device**. Detected face data is **discarded when the app exits** — nothing is persisted.

    ***

    ### Chatbot

    Pepper's chatbot runs **100% on-device**. **No data is sent externally or retained**.

    <Note>Pepper's on-device architecture makes it the most privacy-conservative option in the Plural device family — TTS, face detection, and chatbot all process data locally with no external transmission.</Note>
  </Tab>

  <Tab title="Temi">
    ### Text to Speech (TTS)

    Temi TTS supports two modes:

    * **Google online services** — synthesis via Google's cloud.
    * **On-device Android tools** — local synthesis.

    In both modes, **no data is retained**.

    ***

    ### Speech to Text (STT)

    Temi STT also supports two modes:

    * **Offline — VOSK** ([alphacephei.com/vosk](https://alphacephei.com/vosk/android)) — fully on-device.
    * **Online — Google Cloud Speech** — audio processed by Google's servers.

    In both modes, **data is not retained** by the servers or the device.

    ***

    ### Face Detection and Re-recognition

    Temi performs these detections on-device:

    * Face presence
    * Estimated gender
    * Estimated age
    * Mask detection

    Online name recognition is technically possible but is **not currently in use**.

    All face data is **discarded when the app exits** — nothing is persisted.

    ***

    ### Chatbot

    The Temi chatbot is an online service. The endpoint URL is retrieved from the project's export JSON (currently `https://dev.plural.io/api/chatbot…`).

    **No data is retained** by the chatbot service.

    <Note>Temi and Avatar share the same data-handling architecture. Both support both online and offline STT/TTS modes and discard face data on exit.</Note>
  </Tab>
</Tabs>

## Summary

| Feature                     | Avatar            | Pepper                       | Temi              |
| --------------------------- | ----------------- | ---------------------------- | ----------------- |
| TTS — on device             | Optional          | ✅ Always                     | Optional          |
| TTS — online                | Optional (Google) | ❌                            | Optional (Google) |
| STT — on device             | Optional (VOSK)   | ✅ First attempt              | Optional (VOSK)   |
| STT — online                | Optional (Google) | Fallback (Aldebaran, France) | Optional (Google) |
| Face detection              | On-device         | On-device                    | On-device         |
| Data retained after session | ❌ Never           | ❌ Never                      | ❌ Never           |

If you have data privacy questions specific to your deployment, contact [support@humanizing.com](mailto:support@humanizing.com).