One button. No keyboard.
The voice block renders as an orb, a bar, or a full panel. A visitor taps it, speaks, and the agent answers out loud. There is no text box to find and no account to create before they can talk.
Voice agents
Talk to it. Hear it answer.
One button on the page. A visitor speaks, the agent listens, thinks, and answers out loud in one of six voices — with a visualiser moving in your own brand colours.
No app to install · works in the browser · 14-day free trial
A glowing orb on a published page: tap, speak, and the reply plays back while the aura moves with the audio
6
Voices from gpt-4o-mini-tts
3
Block layouts: orb, bar, panel
20
Speech requests per minute, per IP
4 MB
Upload cap per recording
Four steps between a question and an answer
Capture, transcribe, reply, speak.
1
Capture
A tap opens the microphone. On Chrome the browser recogniser transcribes as the person speaks. Everywhere else the block records the audio and holds it until the person stops.
Web Speech API · MediaRecorder fallback
2
Transcribe
Recorded audio posts to /api/stt as a file. OpenAI whisper-1 returns the text. Uploads are capped at 4 MB, and an optional two-letter language code tells Whisper which language to expect.
POST /api/stt · whisper-1 · 4 MB cap
3
Reply
The transcript goes to the same chat endpoint the text agent uses, so the agent has the same instructions, the same skills and the same memory whether it was typed to or spoken to.
POST /api/chat · one agent, two inputs
4
Speak
The reply text posts to /api/tts and comes back as audio. It plays inline while the visualiser moves with it. If the browser refuses to autoplay, the block shows a tap-to-hear control instead of stalling.
POST /api/tts · mp3 · autoplay-safe
What the voice block gives you
A conversation, not a form.
Six voices from gpt-4o-mini-tts
alloy, echo, fable, onyx, nova, shimmer. Pick one per block, or let a visitor switch it in chat. Where no OpenAI key is set, the agent falls back to Cloudflare Workers AI melotts and still speaks, in one generic voice.
Transcription off Chrome too
Chrome uses the browser recogniser directly. Safari and Firefox record instead and post the audio to /api/stt, which runs OpenAI whisper-1 and returns text. Pass a two-letter language code and the transcript comes back in that language.
A visualiser in your own colours
A glowing aura or moving bars, painted from the same six colour tokens as the rest of the site. Levels come from the live microphone track and from the reply as it plays. Set the visualiser to none and the browser never downloads that code at all.
It never talks over itself
Recognition is switched off while the agent is speaking, so the microphone never transcribes the reply as if it were the visitor. In hands-free mode the microphone reopens the moment the audio ends, and the turn passes back.
Mute, replay, stop
A row of controls under the orb. Mute silences the agent without ending the conversation, replay plays the last answer again, and stop ends the turn. Replay reuses the audio already on the page, so hearing something twice never costs a second generation.
A phone number, if you want one
A separate path connects an ElevenLabs voice agent to a Twilio number, so calls are answered by the same agent and the transcripts land in your inbox. That one needs your own ElevenLabs and Twilio credentials. The in-page block needs neither.
Voice vs. the alternatives
| Feature | ONE | Phone bot platforms | Wire the speech APIs yourself |
|---|---|---|---|
| Voice block you drop on any published page | widget only | you build it | |
| Same agent answers typed and spoken questions | separate bot | you wire it | |
| Audio visualiser using your own brand colours | you build it | ||
| Speech works with no third-party account | |||
| Phone number answered by the agent | yes — separate setup | you build it | |
| Listening with no paid speech keyHonesty: transcription needs an OpenAI key. Chrome covers it with the built-in recogniser; other browsers cannot listen until that key is set. | |||
| Total control of the audio pipelineHonesty: wiring capture, transcription and speech yourself wins here. It is months of work and a bill from every provider you touch. |
Voice is included, not an add-on
Every plan gets the block, the six voices and the visualiser. Annual billing pre-selected.
Starter
Voice block on every page
$500/mo
5M credits/mo
Voice block: orb, bar, panel
Six OpenAI voices
Aura and bars visualiser
Email support
Most popular
Agency
Voice across every client site
$5,000/mo
60M credits/mo
Voice on unlimited client workspaces
Per-workspace voice and language
Hands-free turn taking
Priority support
Scale
Phone line and custom pipeline
$50,000
750M credits/mo
ElevenLabs and Twilio phone agent
Custom speech provider routing
Dedicated success manager
SLA 99.99%
Questions about voice
Hear it before you buy it.
Open the demo, tap the orb, and ask it something. Then put the same block on your own page.
No app to install · no credit card · 14-day free trial