Browser voice (beta)

Browser voice adds a call button to your agent’s widget and test chat, plus a microphone beside the message box for dictation. It listens to one turn, transcribes it through ElevenLabs, generates an answer using your existing Helpnest model and knowledge, and reads that answer aloud. This beta is turn-based; it does not dial phone numbers or provide simultaneous full-duplex conversation.

Enable browser voice

Your server administrator must configure an ElevenLabs API key with Speech to Text and Text to Speech access. Open your agent’s Voice tab, enable the widget button, and save. Only workspace owners and admins can change these settings. Choose from five male and five female voices. Listen to a preview, select a voice, and save. The default is George.

Start a conversation

Reload the widget, click the call button, and choose Start voice call. Allow microphone access, speak, then pause for about a second to send your turn. Send turn submits it manually. Mute pauses microphone capture, Interrupt stops playback so you can speak, and End call stops the microphone. Questions and answers are saved in the same chat and team inbox, including source links.

For speech-to-text only, click the microphone beside the message box, speak, and pause or click Stop dictation. Your words are added to the draft without sending a message. You can edit them and click Send when ready. Dictation shares the workspace’s daily speech allowance.

Request a teammate to end voice and continue in text chat. An agent’s automatic human handoff also continues in text. Teammates do not join a live audio call in this beta.

Limits and usage

Sessions last up to five minutes, with a maximum of 30 seconds per spoken turn. By default, all agents in a workspace share 100 voice-turn attempts per UTC day. The server administrator can configure this allowance. Failed or silent transcription attempts still count because provider usage may already have occurred.

Speech recognition and synthesis use the server’s ElevenLabs API account. AI answers use normal Helpnest credits. These are separate from ElevenAgents hosted call-minute allowances. Your ElevenLabs grant eligibility and rates depend on your account. Long answers may be shortened for audio; the full answer remains in text chat.

Privacy and troubleshooting

Microphone audio is sent to ElevenLabs for transcription, and reply text is sent for speech generation. Helpnest retains the text transcript, not audio recordings. ElevenLabs processing and retention follow your provider account settings; review those settings before enabling voice for customers.

Use HTTPS or localhost and grant microphone permission. For a manually embedded iframe, add allow="microphone; autoplay"; the Helpnest widget script already includes this permission. A parent website’s Permissions-Policy can still block microphone access. If speech playback fails, the answer stays available in the chat.