Chad Bailey from the Pipecat team walks through what's possible with the new Gemini 3 multimodal real-time model: flight search, lodging lookup, Google Search grounding, trip report generation, and a language tutor agent, all in a single voice conversation.

Note: The public string for this model is gemini-3.1-flash-live. The string used in the video is for the Early Access Partner program and is now turned down.

What's covered: Scaffolding a bot with the Pipecat CLI, configuring Gemini 3 with minimal thinking for lower latency, writing system prompts that hold up across long conversations, defining and registering tool calls, enabling Google Search grounding, saving trip reports to disk, and running multiple agents in a single bot file with Pipecat Agents. 

What are you building with Gemini Live API? Drop it in the comments.

Resources:
Gemini Live API overview → https://goo.gle/47vg4Tc 
Get started at pipecat.ai → https://goo.gle/4ch4LAx 
Pipecat examples → https://goo.gle/4uYe93z 

Subscribe to Google for Developers → https://goo.gle/developers 

Speaker: Chad Bailey from the Pipecat
Products Mentioned: Google AI, Gemini

Google for Developers

A walkthrough of building a real-time multimodal voice agent using Pipecat and Google's Gemini 3 model. The demo shows a travel planning bot that handles multi-turn conversations, searches for flights and lodging, uses Google Search grounding, and saves trip reports. Key highlights include Gemini 3's improved instruction following, enhanced tool calling, and multimodal capabilities. The tutorial covers scaffolding with the Pipecat CLI, defining function call handlers, registering tools with the LLM, and running multiple agents (concierge and language tutor) within a single bot file using the Pipecat Agents module.

Build real-time multimodal agents with Gemini and Pipecat