Voice messages and calls → summary, tasks and a voice reply
Upload a recording or speak into the mic. The assistant transcribes it, writes a short summary, pulls out action items (who / what / when) and answers with a voice.
- 1Audio / video
- 2Transcript
- 3Summary
- 4Tasks
- 5Voice reply
LLM
0:00
Say a few sentences, e.g. “Anna will send the report by Friday, I’ll call the client tomorrow”.
3Summary
4Tasks
| What | Who | Due |
|---|
5Voice reply
What’s inside
- Speech-to-text for voice notes, calls and video files (Whisper / gpt-4o-transcribe), timestamps per segment.
- Summary and action items with owner and due date as JSON — ready to push to a CRM, Notion, Jira or a Google Sheet.
- One LLM interface: OpenAI, Anthropic, GigaChat, YandexGPT (or a local model) — switch by config, no code changes.
- Voice reply (TTS) — in a Telegram bot it comes back as a voice message.
- Cost control: rate limit per user, daily budget, maximum audio length.