Real-time Voice-to-Voice Is Here

You can stream microphone audio into an assistant and receive spoken responses over a single WebSocket connection, without wiring together separate speech-to-text, LLM, and text-to-speech services yourself.

Voice is now another interface your agents can use, with the same Backboard infrastructure for memory, tools, and persistent conversation state.

Native Speech-to-Speech Conversations

Connect a microphone stream and receive spoken responses in realtime.

Backboard handles voice interaction over a single WebSocket connection, so you don't need to build and maintain separate STT, LLM, and TTS pipelines.

Choose Your Voice Model

Real-time voice is available across supported models from:

• OpenAI Realtime

• Google Gemini Live

• xAI Grok Voice

Choose the provider and model that fit your application.

Natural Conversation Flow

Voice conversations support the interaction patterns people expect from a realtime conversation, including:

• Streaming audio playback

• Live transcripts

• Automatic turn detection

• Manual turns

• Interruptions

This lets users speak naturally rather than waiting for each interaction to fully finish before starting the next one.

Configure the Voice Experience

You can configure supported provider options such as voice, instructions, speech settings, and turn detection.

Backboard passes those settings through based on what the selected provider and model support.

Start a Conversation Instantly

You can start a realtime conversation without manually creating an assistant or thread first.

Backboard can create them automatically when they are omitted.

You can also reuse an existing thread to continue its saved conversation or reuse an assistant to carry configuration and memory across conversations.

Conversations Persist

Real-time conversations are not limited to the live session.

Final transcripts and tool results are saved to the thread, so you can return to the conversation later.

You can even resume with a different supported provider or model while keeping the conversation history.

Voice Agents Can Use Tools

Real-time voice conversations can also use tools.

That includes your own custom function tools as well as Backboard-hosted capabilities such as:

• Persistent memory

• Web search

• Image generation

• Video generation

This means a voice interface can do more than answer questions. It can take actions and use the same capabilities available to your other Backboard agents.

Python and JavaScript SDKs

The latest SDKs include helpers for realtime connections:

Python: connect_realtime
JavaScript: connectRealtime

These helpers handle streaming audio, real-time events, and returning tool results.

Voice in the Playground

You can now test realtime voice directly in the Backboard dashboard.

Start a microphone conversation, configure the model, hear streamed responses, and follow the live transcript directly in the playground.

Available Now

Real-time voice-to-voice is now available in Backboard.

Explore the documentation → docs.backboard.io

Try it in the Playground → app.backboard.io