v2026.9.5
Release date: September 10, 2026
New Features
Audio replies on the AI Agent
On the Sending tab, with Send messages from the agent on, the prompt can reply with audio using a system voice (ElevenLabs, Minimax, or OpenAI). You do not need TTS nodes in the flow.
What is it for?
- Reply with the brand voice, without building condition, regex, and audio nodes in the flow
- Mirror the customer's audio when they send a voice message
- Control when audio is used (link in the reply and probability on text messages)
- On WhatsApp Official with an auxiliary channel, send audio through WAHA (OGG, more natural sound)
The text is rewritten for speech; the audio arrives with that version as the transcription. If any step fails, the customer gets the original text.
How to use it
- Register a voice in Voices
- Open the AI Agent → Sending
- Turn on Send messages from the agent and Reply with audio
- Choose the voice and the rules (incoming audio, link, probability, and auxiliary channel)
- Remove TTS / audio nodes from the flow if they still exist, to avoid duplicates
Full guide: AI Agents — audio replies
Bug Fixes
Audio and text in the same reply
When the agent replied with audio, the customer also received the text. Now only the audio is sent, with the transcription on the audio itself.
Benefits
- The same audio logic as the flow, without extra nodes
- Brand voice reused from System voices
- On Official, audio can go out through the auxiliary channel (OGG, more natural sound)
- The customer still gets a reply if TTS fails