How do plugNmeet's AI captions, translation and summaries work, and what happens to the data?

plugNmeet includes optional AI features:

• Live captions: speech is turned into text on screen during the session.
• Live translation of captions into other languages, useful for multilingual groups.
• Transcripts you can download after the session (for example as VTT or SRT files).
• Summaries of the session, generated from the transcript.

How they work. These features are not performed inside the video server. When a moderator turns them on, the audio of people speaking is sent to an AI service, which returns text, translations or a summary. That means the AI provider processes participants' voices and what they say.

Which AI provider. It depends on how plugNmeet is run:

• In MynaParrot's plugNmeet service, AI features use Gemini, from Google.
• In a self-hosted installation, the administrator chooses and configures the provider (the project supports OpenAI-compatible providers and is adding self-hosted options).

Encryption. If you enable end-to-end encryption, the server cannot access the audio, so server-side AI features are not available in that session. You choose between E2EE and AI features.

Accuracy. Automatic captions make mistakes, especially with names, technical terms, accents and background noise. Do not treat them as an exact record, and do not rely on them alone where accessibility is a legal requirement.

What to do as an organiser:

1. Tell participants when AI captions, transcripts or summaries are on. 2. Check the data processing terms of the AI provider (location, retention, use for training) and record them in your GDPR documentation. 3. Turn AI features on only in sessions that need them.