Voice / Realtime
Keep your hands on the task. Keep the conversation going.
Let people ask, clarify and change direction out loud. Give your application a spoken interface without stitching every exchange together from separate audio files.
“Which piece goes in first?”
A hands-free momentA conversation, not a voice file
A change of mind is part of the task.
Explore a sample exchange. The person can redirect the conversation; your voice client handles playback and interruptions.
Help me choose a plan for a team of twelve.
What will your team use it for?
Start with a question that moves the task forward.
Mostly answering questions from our customers.
Will all twelve people need access to that workflow?
A follow-up helps make the next answer more useful.
Actually, let's start with our support team. There are four of us.
Let's work with four. What does your support workflow need?
Your client stops the old playback and handles cancellation when the person interrupts.
Worked example · fictional script, not caller transcription · no audio playback or live call.
The same gateway, now speaking
Your gateway already knows. Voice lets people ask.
The realtime session streams through the AI Gateway you already configured—same model, instructions, knowledge and tools.
- ModelRealtime model you chose
- InstructionsSame system prompt
- KnowledgeSame collections
- ToolsSame function calls
No separate voice brain. The spoken exchange reuses the gateway behind your text calls. How gateways work
Your experience, connected
Bring the interface. Connect the conversation.
Your application captures audio, plays the answer and manages interruption. A private backend connects it to the AIVAX realtime session.
- Your audio interface
- Your trusted backend
- AIVAX voice session
Choose the right kind of speech.
Use Realtime for an exchange. Use speech-to-text for a transcript, or text-to-speech when you already have the words.
The agent's spoken replies include text transcripts; incoming caller speech currently does not. Reconnecting starts a new session. Telephony and phone numbers are not included.
Voice integration guideStart with one useful conversation.
Configure an AI Gateway and connect a backend audio client. Choose a task whose outcome you can recognize and evaluate.
Understand the cost before scaling.
Voice follows the selected realtime model's token rates, not a universal per-minute price. Keep the private API key on your backend, never in the browser.
Voice billing and limits