LLM

Gemini Live speech-to-speech for Bolna AI voice agents

Gemini Live runs a Bolna voice agent on Google's native-audio speech-to-speech model, with 30 prebuilt voices and session resumption for calls past 15 minutes.

Book a Demo Bolna Docs

llmspeech-to-speechgeminigooglerealtimenative-audio

Official Documentation

At a glance

How Gemini Live fits in the stack

Best for

Reasoning, policy handling, tool use, and response generation.

Use this layer when

You want to tune quality, cost, latency, or function-calling behavior.

Connects to

Transcribed caller input upstream and voice or actions downstream.

Voice Stack

The LLM is the reasoning layer

This provider decides how the agent interprets intent, selects next actions, calls tools, and formulates responses. It is the decision engine inside the voice stack.

TelephonyPhone Network

→

STTListener

→

LLMReasoning

→

TTSVoice

→

ToolsActions

Last updated 25 October 2026

What is Gemini Live used for in a Bolna voice agent?

Gemini Live lets a Bolna agent run on one Google speech-to-speech model that hears the caller, calls your tools and speaks the reply itself, replacing the separate transcriber, LLM and synthesizer. Bolna uses Gemini's native audio models, where the model produces speech directly rather than handing text to a voice. Long calls keep their thread through session resumption and context compression.

How do I set up Gemini Live with Bolna?

In Agent Studio, switch the agent's Voice engine from Cascaded to Realtime (Beta).
Use the Model tab, which replaces the Intelligence and Languages tabs, to pick the model and voice.
In the API, set agent_type to realtime_agent, add tools_config.s2s with provider set to gemini_live, and set the toolchain pipeline to [["s2s"]].
Choose a model, such as gemini-3.1-flash-live-preview, and a voice, such as Kore.
Set language as a BCP-47 code, for example en-US or hi-IN.
Leave enable_session_resumption and enable_context_compression on for any call that can run past 15 minutes.

Every field is listed in the Gemini Live setup guide in the Bolna docs.

Which Gemini Live models does Bolna support?

Model	Notes
`gemini-3.1-flash-live-preview`	Default. Low-latency audio to audio for real-time dialogue
`gemini-2.5-flash-native-audio-preview-12-2025`	Native audio reasoning, with video support

Google lists the Live API models as preview, with tighter rate limits than its generally available models. Bolna offers 30 prebuilt voices, including Kore, Puck, Charon, Aoede, Fenrir, Leda, Orus and Zephyr.

What can you build with Gemini Live and Bolna?

Long support and counselling calls. Session resumption restores the session when Gemini rotates it at around 15 minutes, and context compression keeps the call under the model's limit. Watch cost, since input tokens grow each turn.

Agents for callers who pause. Set end_sensitivity to END_SENSITIVITY_LOW and raise vad_silence_duration_ms towards 800ms. Google's guidance puts the usable band at 500 to 800ms.

When to keep the classic pipeline. Speech-to-speech does not support knowledge bases, graph agents or multilingual configuration. For those, run the Gemini LLM with a transcriber such as Gemini speech-to-text and a separate voice.

Comparing realtime providers. OpenAI Realtime is the other speech-to-speech option on Bolna, with semantic turn detection and reasoning models.

FAQ

Gemini Live on Bolna: frequently asked questions

Yes. Create a speech-to-speech agent with an `s2s` block whose provider is `gemini_live`. The default model is `gemini-3.1-flash-live-preview` with the `Kore` voice.

Yes. Gemini caps an audio session at roughly 15 minutes, and with `enable_session_resumption` on, Bolna restores the session so the caller notices nothing.

No. Knowledge base agents, graph agents, multilingual configuration, per-call voice overrides and backchanneling are not available on speech-to-speech agents.

Gemini resends the conversation each turn, so input tokens grow as the call goes on. Cost scales with call length faster than on a transcriber and LLM pipeline.

Other providers in this layer

LLM

Anthropic

Anthropic Claude writes your Bolna voice agent's replies. Use claude-sonnet-5 for most agents, or claude-haiku-4-5 for faster, high-volume calls.

LLM

Azure OpenAI

Azure OpenAI runs GPT models for your Bolna voice agent inside your own Azure account, for teams that need regional data residency and private networking.

LLM

DeepSeek

DeepSeek V4 runs the conversation in a Bolna voice agent at low cost per token. deepseek-v4-flash suits high-volume calls, with a 1M token context window.

Browse this layer

Keep exploring the voice stack

Browse

Large Language Models

The LLM is the brain of your voice agent. It understands what callers say and decides how to respond. Bolna lets you switch between models like GPT-5.4 mini, Claude and DeepSeek by changing the provider and model in your agent LLM settings, so you can tune for speed, cost or reasoning depth.

Browse

Telephony

Telephony providers connect your voice agents to the phone network so they can make and receive real calls. Bolna supports managed integrations with major carriers as well as bring-your-own-carrier via SIP trunking, giving you full control over call routing, number provisioning, and cost.

Browse

Speech-to-Text

Speech-to-text converts what callers say into text that your LLM can process. Transcription accuracy and latency directly affect how natural a conversation feels. Bolna supports streaming STT providers optimized for telephony audio, including specialized models for Indian languages.

Browse

Text-to-Speech

Text-to-speech turns your agent responses into spoken audio. Voice quality shapes how callers perceive your brand. Flat, robotic speech kills trust while natural, expressive voices build it. Bolna integrates with the fastest TTS providers so responses sound human and arrive without awkward pauses.

Browse

Tools & Workflows

Tools let your voice agents take action during a call, not just talk. Book a calendar slot, look up an order through your own API, push a lead into your CRM, or trigger a multi-step automation in Zapier. These integrations turn voice agents from answering machines into workflow engines.

See where Gemini Live fits in your production workflow

Use the demo to walk through provider selection, stack tradeoffs, and the exact workflow you want Bolna to automate.

Book a Demo Read Bolna Docs