How do you make a clip your app can play?
Open Text to Speech in the ElevenLabs dashboard. Enter a short script, choose a voice you can use and download the result. ElevenLabs documents MP3 and WAV downloads from generation history.
Start with a fictional line such as “The reading room closes at four on Thursday.” Listen for the room, the day and the time. Put the approved words next to the play button so the visitor can get the information without sound.
When everyone hears the same clip, your app can serve the saved file. When each visitor needs a different line read aloud, your app needs a connection to request and play the new audio. Keep the credential on your server, out of the browser code.
How can your app get text from a recording or microphone?
Send an existing audio file for a transcript, or use the live transcription integration when the person needs to see words as they speak.
The browser integration uses a temporary, single-use token from your server to start a live session. Decide who can start one, ask for microphone access and show when listening is active. The integration can return partial words and then a settled segment. Let the person correct names, times and numbers before saving or acting on the text.
Keep an editable text field when someone declines microphone access or the session drops. For uploaded files, decide who can read the original and the returned transcript and what the app will keep.
What do you need for a public or private voice assistant?
Set up a conversational agent in the dashboard or through the API. Give it a small set of facts and try a reply in a separate test agent before connecting it to an app.
Just before people can use the assistant on the web or by phone, tell them it is AI. Explain that ElevenLabs, its affiliates and its service providers process the conversation, including the provider of any model you select. If recording or conversation retention is on, say what will be recorded and what will be kept. Get any required consent before continuing and keep a record when required. Give anyone who declines a way to get help by text or from a person, and check that it works.
The basic website widget instructions use a public agent. For an assistant for signed-in customers, ElevenLabs documents an SDK, a set of tools your developer can use to connect the app. Your server checks the customer’s access first and requests a temporary connection link. The browser uses that link to talk to the agent, and the private API key stays on your server.
A phone line needs its own setup. The documented Twilio integration supports incoming calls with a provisioned Twilio number. A verified caller ID in that integration is for outbound calls only. Keep the first test away from real calls, customer records and tools that can change anything.
Which parts of ElevenLabs usage affect the bill?
Estimate the workflow you will run. Speech generation, transcription and agent conversations use different units. Check the exact model and plan, including the difference between using the browser tool and calling it from your app.
If you use prepaid usage, check what the plan includes, whether automatic top-up is on and who can change it. The service can pause when the available balance runs out and a top-up does not happen. Give people another way to finish their task if voice is unavailable.
| What you use | What to count |
|---|---|
| Speech generation through the API | Text characters, including spaces and punctuation. The rate depends on the model and pricing path. |
| Recorded or live transcription | The duration of the audio sent for transcription. Check which mode and options the app will use. |
| Agent conversation | For voice-only use, count time connected and check the treatment of silence. Text-only conversations use a per-message charge. Add the language-model charge and any external carrier charge that applies. |
What can you keep if you change tools?
Access to hosted files can end after cancellation. Download the clips you need beforehand and save their approved words alongside them.
ElevenLabs grants commercial use for qualifying content made during a paid subscription, subject to input rights and terms. This excludes Beta Services. Check that the voice, script and intended use are covered before publishing a clip for a business.
For an agent, the API can retrieve one conversation’s details, including the transcript, and its audio if a recording was saved. It can also retrieve the agent’s configuration. Keep your own copy of the instructions and sample conversations. Check separately whether a custom voice, connected services or call routing can be moved before relying on a transfer.
Which recordings and transcripts will the agent keep?
Check the agent’s recording and retention settings before anyone speaks into it. ElevenLabs lets you choose whether it saves call audio and how long conversation data stays available.
Look at audio and transcripts separately. The audio-saving control and the period for keeping transcripts are separate settings. When changing a retention period, check whether it applies to previous conversations or only new ones.
For the test agent, write down what you expect to see after a conversation ends and which parts, if any, your own app or a connected service would keep.