Post Sessions

Google Meet or Microsoft Teams meeting link

AI agent ID. Omit (or null) for TRANSCRIBE-ONLY: the bot joins and transcribes into the thread but never speaks and runs no agentic session. A ws-0 template id is accepted and forked into the workspace on use.

Screen-share vision mode. None → default-derive from the agent's vision_enabled flag (enabled → 'continuous_0_3fps', else 'off'). 'off' = no vision; 'on_demand' = the agent can call vision.query; 'continuous_0_3fps' = every agent turn carries a low-detail ambient screen frame + vision.query for fine detail.

Task brief for the agent once it joins (e.g. 'greet everyone and present the overview deck'). Woven into the voice system prompt.

If true, the agent starts talking as soon as it joins (greets + begins the task) instead of waiting for a wake-word.

If true, pre-arm translator mode (auto-target language) on join with zero verbal trigger. Forces meet_audio_path ∈ {framework, hybrid} so the original-speaker track exists for listener audio path.

Primary target language spoken INTO the Meet call (e.g. 'en', 'es', 'fr'). Used with translation_mode=True.

Speech-recognition provider for THIS session: 'deepgram' or 'gladia'. Omit to use the agent's setting, then the workspace default. Gladia covers ~115 languages including Uzbek, Kazakh and Tajik, which Deepgram does not.

Recognition model within the provider, e.g. 'flux' or 'nova-3' for Deepgram. Omit to use the agent's setting, then the workspace default.

Spoken language of the meeting (any ISO code: 'en', 'ru', 'zh', ...). When set, speech recognition runs in that language's dedicated mode for better accuracy. Omit when participants may speak multiple languages — recognition then autodetects. Independent of translation_mode: improves plain transcription too.

Translator VOICE provider (the TTS the translator speaks with). One of the translation-capable providers (cartesia, openai, deepgram, ...). Only with translation_mode. Omit for the workspace default translation voice.

Specific voice id for tts_provider (e.g. 'nova' for openai). Omit for the provider's default voice.

Silent translator dispatch: the translator never speaks into the Meet call (participants hear no bot voice). Translation audio + subtitles remain available in the app (translator-tts tracks / listen overlay). Only meaningful with translation_mode=True.

Phrase or word to trigger return to original language during translation. e.g. 'Back to English'. Optional.

Audio routing strategy: 'direct' (bot input bypasses framework), 'framework' (bot hears all participants via SFU), 'hybrid' (bot hears and speaks via framework, guests routed direct). None → default-derive. Rejected when translation_mode=True (must be framework or hybrid).

Unique session ID

Thread ID (voice_sessions.thread_id cast to int)

Channel reference (meet:<code>)

Session status (always 'joining')

Public zero-auth translation page URL (/live/{token}) — set only for translation_mode dispatches. Share or QR it; viewers get subtitles + translated audio, never meeting access.