Can Your Agent Hear You Now? Building Live Voice Agents with Gemini — Thor Schaeff

Can you hear me? Perfectly. Can you hear me? Perfectly. Hello everyone. Well, I welcome Hello everyone. Well, I welcome Hello everyone. Well, I welcome those few who those few who those few who came here. Sorry came here. Sorry that the schedule that the schedule that the schedule was constantly changing. I'm was constantly changing. I'm was constantly changing. I'm sorry you didn't sorry you didn't sorry you didn't find the quiet find the quiet find the quiet workspace you were looking for. workspace you were looking for. OK. We will look at OK. We will look at some agents with the some agents with the some agents with the Gemini API. Before we Gemini API. Before we Gemini API. Before we start, let's start, let's start, let's listen to this. listen to this. Welcome everyone to the Welcome everyone to the greatest stage greatest stage greatest stage on Earth. You are standing in the on Earth. You are standing in the on Earth. You are standing in the heart of the heart of the heart of the largest largest largest exhibition of exhibition of exhibition of artificial intelligence engineers in history. artificial intelligence engineers in history. Let's build the future Let's build the future together, okay? Does anyone together, okay? Does anyone together, okay? Does anyone know what model this is? know what model this is? Yes, you, you, you, you are Yes, you, you, you, you are not allowed. Anyone not allowed. Anyone not allowed. Anyone else? Does anyone know what else? Does anyone know what else? Does anyone know what model this is? No? Any guesses model this is? No? Any guesses ? It starts with ? It starts with ? It starts with the letter "G". Gemini? Yes, the letter "G". Gemini? Yes, the letter "G". Gemini? Yes, it's a Gemini model, but which one it's a Gemini model, but which one it's a Gemini model, but which one exactly? No, not "life". This is exactly? No, not "life". This is exactly? No, not "life". This is wrong. Yes, this is wrong. Yes, this is wrong. Yes, this is Gemini TTS. Is that what you wanted to Gemini TTS. Is that what you wanted to Gemini TTS. Is that what you wanted to say? No? OK.

say? No? OK. So, Gemini text-to-speech is a So, Gemini text-to-speech is a truly incredible truly incredible truly incredible model for expressive model for expressive model for expressive speech. It has speech. It has speech. It has built-in audio tags built-in audio tags built-in audio tags that you can manage. that you can manage. that you can manage. You can You can You can stream the broadcast stream the broadcast stream the broadcast . . . Now it has much Now it has much Now it has much less latency than less latency than less latency than before. So, if you before. So, if you before. So, if you are looking for a are looking for a are looking for a speech synthesis solution, Gemini speech synthesis solution, Gemini speech synthesis solution, Gemini TTS is a great model. TTS is a great model. Now let's Now let's see what we're see what we're see what we're going to do today. going to do today. First of all, my First of all, my name is Thor. I work name is Thor. I work name is Thor. I work on on on developer experience at Google developer experience at Google developer experience at Google DeepMind. This mainly DeepMind. This mainly DeepMind. This mainly includes Gemini API and Google includes Gemini API and Google includes Gemini API and Google AI Studio. So, if you AI Studio. So, if you AI Studio. So, if you are using are using are using Gemini models via the API, Gemini models via the API, Gemini models via the API, let me know let me know let me know if you have any feedback. I'm if you have any feedback. I'm if you have any feedback. I'm also joined by my also joined by my also joined by my colleague Philip, who colleague Philip, who colleague Philip, who works on works on works on managed agents managed agents managed agents in the API. So, if you in the API. So, if you in the API. So, if you use them, use them, use them, let us know too. This is a let us know too. This is a let us know too. This is a small announcement of small announcement of small announcement of our our our workshop tomorrow. Tomorrow we will workshop tomorrow. Tomorrow we will workshop tomorrow. Tomorrow we will look at the Gemini API, look at the Gemini API, look at the Gemini API, focusing on " focusing on " live" agents, live" agents, live" agents, conversational agents, conversational agents, real-time conversational agents, and real-time conversational agents, and multimodal multimodal real-time conversational agents. We're real-time conversational agents. We're just trying to just trying to just trying to squeeze everything in there.

squeeze everything in there. In fact, we can In fact, we can consider many consider many consider many things that have already come out. things that have already come out. In particular, we want to In particular, we want to look at look at look at the multimodality of the multimodality of the multimodality of Gemini and Gen Media models. Gemini and Gen Media models. This slide is This slide is no longer relevant because no longer relevant because no longer relevant because this morning we this morning we this morning we released a new released a new released a new model model model called...anyone called...anyone called...anyone know? What model did we know? What model did we know? What model did we release? We release? We release? We released Omni Flash, a released Omni Flash, a released Omni Flash, a model for creating model for creating model for creating and editing videos. and editing videos. We also released Nano We also released Nano Banana Lights. So, Nano Banana Banana Lights. So, Nano Banana Banana Lights. So, Nano Banana Lights is a really Lights is a really Lights is a really interesting model interesting model interesting model because it significantly because it significantly because it significantly reduces latency reduces latency reduces latency when creating and when creating and when creating and editing editing editing images. So, we're images. So, we're images. So, we're going to do a few going to do a few going to do a few demonstrations here. One of demonstrations here. One of demonstrations here. One of them is an example that them is an example that them is an example that I call the Gemini Nano Banana I call the Gemini Nano Banana I call the Gemini Nano Banana Story Canvas. I just Story Canvas. I just Story Canvas. I just wrote the code for this wrote the code for this , and I'm having , and I'm having , and I'm having some issues with the some issues with the some issues with the mouse here. But we can mouse here. But we can mouse here. But we can come up with a story come up with a story come up with a story together with Gemini Life and together with Gemini Life and together with Gemini Life and then use Nano then use Nano then use Nano Banana Lights to Banana Lights to Banana Lights to illustrate it illustrate it illustrate it for us. It's a bit like for us. It's a bit like improvised theater . Does anyone have a story . Does anyone have a story . Does anyone have a story we want to we want to we want to start with? Where start with? Where start with? Where are we? Somewhere...

are we? Somewhere... Okay, let's say... We're in Okay, let's say... We're in Okay, let's say... We're in Taipei in the 1980s. The Taipei in the 1980s. The neon lights of neon lights of Taipei's night market Taipei's night market Taipei's night market pulsed with an almost pulsed with an almost pulsed with an almost magical energy, magical energy, magical energy, reflecting off the reflecting off the reflecting off the slick asphalt and slick asphalt and slick asphalt and chrome of the scooters that chrome of the scooters that chrome of the scooters that passed by. The rain had passed by. The rain had passed by. The rain had just stopped, leaving a just stopped, leaving a just stopped, leaving a damp haze damp haze damp haze hanging in the air. hanging in the air. hanging in the air. Amidst the crowd of Amidst the crowd of Amidst the crowd of buyers and sellers, a buyers and sellers, a buyers and sellers, a young woman young woman young woman clutched a clutched a clutched a small, strangely small, strangely small, strangely glowing object tightly beneath her jacket. glowing object tightly beneath her jacket. glowing object tightly beneath her jacket. Something very exciting Something very exciting Something very exciting happened in Taipei in the 1980s. A happened in Taipei in the 1980s. A happened in Taipei in the 1980s. A new type of drink has appeared new type of drink has appeared new type of drink has appeared . . She stopped at a She stopped at a stall where a man was stall where a man was stall where a man was vigorously shaking a vigorously shaking a vigorously shaking a metal cup, metal cup, metal cup, creating a frothy creating a frothy creating a frothy milk mixture. He milk mixture. He milk mixture. He handed the cup to handed the cup to handed the cup to the customer, who took a the customer, who took a the customer, who took a long sip, long sip, long sip, eyes wide and eyes wide and eyes wide and savoring the strange savoring the strange savoring the strange tapioca balls at the bottom tapioca balls at the bottom . This was the birth of . This was the birth of . This was the birth of bubble tea, and its bubble tea, and its bubble tea, and its sweet, chewy sweet, chewy sweet, chewy magic spread magic spread magic spread like wildfire. like wildfire. So, you can So, you can see here the result of see here the result of see here the result of calling the tool.

calling the tool. calling the tool. Gemini Life receives a Gemini Life receives a Gemini Life receives a tool challenge. And then tool challenge. And then tool challenge. And then it uses the it uses the it uses the Gemini Interactions API to Gemini Interactions API to Gemini Interactions API to create create create images on the fly. images on the fly. images on the fly. Right now it still Right now it still Right now it still takes about two to takes about two to three seconds to three seconds to three seconds to directly directly directly generate an image generate an image . But it's pretty . But it's pretty . But it's pretty impressively fast. You impressively fast. You impressively fast. You can look at the can look at the can look at the code, we're code, we're code, we're using the using the using the Interactions API, and it's interesting Interactions API, and it's interesting Interactions API, and it's interesting because it's a stateful API, because it's a stateful API, because it's a stateful API, which means we which means we which means we can can can combine our combine our combine our interactions very easily. As you can see, the interactions very easily. As you can see, the interactions very easily. As you can see, the first first first image generation we did did image generation we did did image generation we did did not have a not have a not have a previous previous previous interaction ID. But for the interaction ID. But for the interaction ID. But for the next generation, next generation, next generation, we pass this we pass this we pass this previous previous previous interaction ID to the API, which interaction ID to the API, which interaction ID to the API, which means we have means we have means we have the context of the previous the context of the previous the context of the previous interaction and can interaction and can interaction and can continue that in the continue that in the continue that in the generation. So what this generation. So what this generation. So what this means specifically for means specifically for means specifically for our demo is that our demo is that our demo is that you can see that you can see that you can see that our first, you know, our first, you know, our first, you know, interaction interaction interaction generated generated generated an image with a very an image with a very an image with a very specific specific specific visual feel visual feel , and also with the , and also with the , and also with the characters, characters, characters, right? And so we can right? And so we can right? And so we can continue this continue this continue this interaction, and we interaction, and we interaction, and we can maintain the same can maintain the same can maintain the same consistency across these consistency across these consistency across these generations. So if generations. So if generations. So if you haven't tried the you haven't tried the you haven't tried the Interaction API yet, it's our Interaction API yet, it's our Interaction API yet, it's our unified API for unified API for unified API for working with all of working with all of working with all of Google DeepMind's models as well as Google DeepMind's models as well as Google DeepMind's models as well as agents, so be agents, so be agents, so be sure to give it a sure to give it a sure to give it a try. This is a really

try. This is a really try. This is a really , really cool , really cool , really cool way to interact with way to interact with way to interact with everything that Google DeepMind has. everything that Google DeepMind has. OK. So, it was, OK. So, it was, you know, a lot of events you know, a lot of events you know, a lot of events that happened. So, this that happened. So, this that happened. So, this was from Gemini media, we was from Gemini media, we was from Gemini media, we released the Nano Banana light released the Nano Banana light released the Nano Banana light this morning. this morning. this morning. Now, we've also Now, we've also Now, we've also recently released recently released recently released our live translation model on the audio side our live translation model on the audio side . Has anyone . Has anyone tried tried tried Gemini 3.5 live translation yet? Yeah, Gemini 3.5 live translation yet? Yeah, Gemini 3.5 live translation yet? Yeah, okay, a few people. okay, a few people. okay, a few people. This is just wonderful. So what This is just wonderful. So what This is just wonderful. So what we can do is this we can do is this we can do is this : we can : we can : we can try this too. try this too. try this too. We can show a We can show a We can show a little little little demonstration. So demonstration. So demonstration. So what I'm going to what I'm going to what I'm going to do here is create a do here is create a do here is create a live live live translation session, and I want translation session, and I want translation session, and I want you to scan you to scan you to scan this QR code on your this QR code on your this QR code on your phone. And here's what it phone. And here's what it phone. And here's what it does: if I does: if I does: if I turn on my turn on my turn on my microphone right now, it microphone right now, it microphone right now, it will transmit my will transmit my will transmit my speech here, into this session speech here, into this session . And now, if you . And now, if you . And now, if you go in and select go in and select go in and select your preferred language, we your preferred language, we your preferred language, we see that we get see that we get see that we get German, so someone German, so someone German, so someone is listening to me is listening to me is listening to me speak German right now.

speak German right now. So if you're listening to this right now So if you're listening to this right now , you , you , you probably need probably need probably need headphones, so if headphones, so if headphones, so if you have you have you have noise-canceling headphones, you noise-canceling headphones, you noise-canceling headphones, you can listen to this can listen to this can listen to this conversation in your language. conversation in your language. Gemini live translate is a Gemini live translate is a truly incredible truly incredible truly incredible model. This is model. This is model. This is almost real- almost real- almost real- time translation. So, as I time translation. So, as I time translation. So, as I speak, once speak, once speak, once the model recognizes the model recognizes the model recognizes the context of what the context of what the context of what I'm saying, it will I'm saying, it will I'm saying, it will start translating in start translating in start translating in real time into real time into real time into your desired language. your desired language. I see someone I see someone is listening there. Are you getting is listening there. Are you getting is listening there. Are you getting sound? Can you hear us? sound? Can you hear us? sound? Can you hear us? Yes? Does it sound good? Yes? Does it sound good? Yes? Does it sound good? Fantasy. See? Fantasy. See? Fantasy. See? So it's really So it's really So it's really exciting because, exciting because, exciting because, you know, at a you know, at a you know, at a conference like this, conference like this, conference like this, we obviously have a very, we obviously have a very, we obviously have a very, very international very international very international audience. Therefore, I audience. Therefore, I audience. Therefore, I can speak my can speak my can speak my preferred language. preferred language. preferred language. For example, I could For example, I could For example, I could switch to switch to switch to German now. This is the German now. This is the German now. This is the true true true power of Gemini live power of Gemini live power of Gemini live translate. This is truly an translate. This is truly an translate. This is truly an incredible model.

incredible model. It automatically It automatically detects the language detects the language detects the language the speaker is speaking and the speaker is speaking and the speaker is speaking and then translates it in then translates it in then translates it in real time into real time into real time into our desired language. our desired language. So, we have a So, we have a very international very international very international audience today. Who here audience today. Who here audience today. Who here is listening in Chinese? Are you is listening in Chinese? Are you is listening in Chinese? Are you listening in Chinese? listening in Chinese? Good. I didn't Good. I didn't understand that, but what are you understand that, but what are you understand that, but what are you saying there? OK. saying there? OK. saying there? OK. So, this is Gemini Live Translate. So, this is Gemini Live Translate. If you want If you want to try it, you can to try it, you can to try it, you can add it to your add it to your add it to your apps right now. apps right now. apps right now. This is available through the Gemini This is available through the Gemini This is available through the Gemini API. So be sure API. So be sure API. So be sure to try it. This is a truly to try it. This is a truly to try it. This is a truly incredible model and it would be incredible model and it would be incredible model and it would be interesting interesting interesting to see what you create with it to see what you create with it to see what you create with it . Now, . Now, . Now, obviously we're talking obviously we're talking obviously we're talking about agents, right? about agents, right? about agents, right? Can your agents... Can your agents... What is the title of this What is the title of this What is the title of this report? Gemini report? Gemini report? Gemini came up with this name. came up with this name. came up with this name. So, can your agent hear me So, can your agent hear me So, can your agent hear me now? I think that's the now? I think that's the now? I think that's the name. So, our name. So, our name. So, our newest model for newest model for newest model for real- real- real- time conversations is called Gemini time conversations is called Gemini time conversations is called Gemini 3.1 Flash Live. If you have 3.1 Flash Live. If you have 3.1 Flash Live. If you have previously chatted previously chatted previously chatted with Gemini on the Gemini app, with Gemini on the Gemini app, with Gemini on the Gemini app, this is the model this is the model this is the model you are talking to.

you are talking to. you are talking to. So we can do a So we can do a So we can do a little little little demonstration. This demonstration. This demonstration. This demo was demo was demo was created in real created in real created in real time by the Gemini model in time by the Gemini model in time by the Gemini model in Google AI Studio. If you Google AI Studio. If you Google AI Studio. If you remember the remember the remember the good old days when you good old days when you good old days when you listened to the radio, not so listened to the radio, not so listened to the radio, not so long ago, you could long ago, you could long ago, you could call call call the radio host and the radio host and the radio host and ask for a song, ask for a song, ask for a song, right? You could right? You could right? You could order a song. order a song. order a song. The idea is the same here. We The idea is the same here. We The idea is the same here. We use Gemini use Gemini use Gemini Live as our radio Live as our radio DJ. We provide DJ. We provide DJ. We provide Gemini Live with a Gemini Live with a Gemini Live with a call function via the Gemini API call function via the Gemini API call function via the Gemini API to then to then to then use Lyria 3, use Lyria 3, use Lyria 3, our our our music generation model, to music generation model, to music generation model, to create music on create music on create music on demand. Okay, demand. Okay, demand. Okay, let's try. We let's try. We let's try. We move from move from move from one live one live one live demonstration to demonstration to demonstration to another. Let's take another. Let's take another. Let's take a look. a look. a look. Greetings. You are now live on the Greetings. You are now live on the live live jukebox. What hit will we have jukebox. What hit will we have jukebox. What hit will we have for you today for you today for you today , friend? , friend? , friend? To make this track To make this track To make this track right, I right, I right, I need a few need a few need a few more ideas. What more ideas. What more ideas. What genre will we choose?

genre will we choose? genre will we choose? Energetic synth-pop, Energetic synth-pop, Energetic synth-pop, smooth jazz, or something smooth jazz, or something smooth jazz, or something more driving? And more driving? And more driving? And what mood should the what mood should the what mood should the text be in? Give us text be in? Give us text be in? Give us some details. some details. some details. So, can we have a really So, can we have a really So, can we have a really energetic reggaeton energetic reggaeton energetic reggaeton about the about the about the AI ​​developer fair here in AI ​​developer fair here in AI ​​developer fair here in San Francisco, San Francisco, San Francisco, please? please? Oh, that sounds like a Oh, that sounds like a real hit, buddy. real hit, buddy. real hit, buddy. Energetic reggaeton Energetic reggaeton Energetic reggaeton about the about the about the AI ​​developer fair, AI ​​developer fair, AI ​​developer fair, brilliant. So, to brilliant. So, to brilliant. So, to make this a truly make this a truly make this a truly top hit, are there top hit, are there top hit, are there any specific any specific any specific technologies or names technologies or names technologies or names you want to add? you want to add? you want to add? Are we just Are we just Are we just focusing on focusing on focusing on the atmosphere of a big the atmosphere of a big the atmosphere of a big festival? festival? festival? Yes, just surprise Yes, just surprise Yes, just surprise me, please. me, please. Agreed, my friend . Get ready for a . Get ready for a . Get ready for a grand grand grand musical surprise. musical surprise. musical surprise. I'm generating this track I'm generating this track I'm generating this track right now. I'm mixing a right now. I'm mixing a right now. I'm mixing a reggaeton beat with a techno reggaeton beat with a techno vibe for the vibe for the vibe for the World's Fair World's Fair . Watch out for this, it'll . Watch out for this, it'll . Watch out for this, it'll just rip your just rip your just rip your headphones off. Boom! headphones off. Boom!

Yeah, that was pretty Yeah, that was pretty cool. You also cool. You also cool. You also switched to several switched to several switched to several languages. Good job. languages. Good job. languages. Good job. Thank you. Thank you. Thank you. Haha, thanks, buddy. I'm glad Haha, thanks, buddy. I'm glad Haha, thanks, buddy. I'm glad you like it you like it you like it . Keep . Keep . Keep sharing great sharing great sharing great ideas, this ideas, this ideas, this jukebox never jukebox never jukebox never sleeps. What will we sleeps. What will we sleeps. What will we put next? put next? put next? I feel I feel I feel good. Um, yes, you good. Um, yes, you good. Um, yes, you can try it can try it can try it yourself. This is a demo yourself. This is a demo yourself. This is a demo created in Google AI Studio. created in Google AI Studio. So, if you want, So, if you want, you can fork you can fork you can fork AI Studio and continue on AI Studio and continue on AI Studio and continue on your own. I know your own. I know your own. I know this app may this app may this app may not be entirely not be entirely not be entirely useful for your useful for your useful for your business context business context business context if that's what you're looking for if that's what you're looking for if that's what you're looking for . But I think that . But I think that . But I think that individual components individual components individual components will definitely come in will definitely come in will definitely come in handy. So if you're handy. So if you're handy. So if you're building building building real- real- real- time conversational agents, you want them to time conversational agents, you want them to time conversational agents, you want them to be multilingual be multilingual be multilingual and have and have and have tool-invocation capabilities tool-invocation capabilities . In this example, we . In this example, we . In this example, we simply instructed the agent simply instructed the agent simply instructed the agent to generate to generate to generate music using music using music using Lyra 3 on demand. Of course, Lyra 3 on demand. Of course, Lyra 3 on demand. Of course, it could also be a it could also be a it could also be a call to your CRM to call to your CRM to call to your CRM to get get get customer information or customer information or customer information or create a ticket in the create a ticket in the create a ticket in the backend. So the main thing backend. So the main thing backend. So the main thing I want you to I want you to I want you to take away from here is the take away from here is the take away from here is the experience of creating experience of creating experience of creating multimodal multimodal multimodal conversations in real conversations in real conversations in real time. The way to work with time. The way to work with time. The way to work with this model, Gemini 3.1 this model, Gemini 3.1 this model, Gemini 3.1 Flash Live, is to Flash Live, is to Flash Live, is to use the Live API. Live use the Live API. Live use the Live API. Live API is a API is a API is a low-latency

low-latency low-latency interface for working interface for working interface for working with voice and video in with voice and video in with voice and video in real time. You real time. You real time. You can create can create can create multimodal multimodal multimodal interactive interactive interactive interactions interactions interactions live. This means live. This means live. This means you have a you have a you have a WebSocket API where you WebSocket API where you WebSocket API where you create a session create a session create a session through which you send through which you send through which you send real-time real-time real-time text, audio, and video, text, audio, and video, text, audio, and video, and the model responds with and the model responds with and the model responds with audio snippets. audio snippets. audio snippets. You You You can broadcast these audio clips in can broadcast these audio clips in can broadcast these audio clips in your application. your application. In addition to the audio, you also In addition to the audio, you also get a transcript of get a transcript of get a transcript of what was said. If you've what was said. If you've what was said. If you've worked with worked with worked with voice agents before, voice agents before, voice agents before, this is a little this is a little this is a little different from the different from the different from the cascade pipeline cascade pipeline cascade pipeline because it's a native because it's a native because it's a native audio model. So, audio model. So, audio model. So, the model itself does not work the model itself does not work the model itself does not work through text, through text, through text, instead the intelligence instead the intelligence instead the intelligence is built is built is built directly into directly into directly into the audio model. So the audio model. So the audio model. So the audio model can the audio model can the audio model can understand speech understand speech understand speech naturally and in its native naturally and in its native naturally and in its native language, and then language, and then language, and then respond in respond in any of the 90 any of the 90 any of the 90 supported languages.

supported languages. This is what allows us to switch between to switch between different languages ​​very smoothly. So, different languages ​​very smoothly. So, different languages ​​very smoothly. So, the key features: the key features: the key features: multilingual multilingual multilingual support, currently support, currently 90 different languages ​​are supported in the preview. We 90 different languages ​​are supported in the preview. We can intervene, can intervene, can intervene, interrupt the model if interrupt the model if interrupt the model if it talks it talks it talks too long. We too long. We too long. We can intervene and can intervene and can intervene and kill her. We kill her. We kill her. We can provide her with can provide her with can provide her with tools, and we also tools, and we also tools, and we also have audio have audio transcription. If you transcription. If you transcription. If you build build build real- real- real- time conversational apps or time conversational apps or time conversational apps or multimodal multimodal multimodal apps, you know that apps, you know that apps, you know that scaling scaling scaling real-time media is real-time media is real-time media is quite challenging. We quite challenging. We quite challenging. We have a great line of have a great line of have a great line of partners. LifeKit, PipeKit— partners. LifeKit, PipeKit— they're all here at the they're all here at the they're all here at the show now. I show now. I show now. I recommend you recommend you recommend you visit them. They visit them. They visit them. They do some really cool do some really cool do some really cool stuff that will help stuff that will help stuff that will help you get you get you get started with the Gemini Live API quickly. started with the Gemini Live API quickly. OK. So if you're OK. So if you're thinking, "Okay, that thinking, "Okay, that thinking, "Okay, that sounds incredible."

sounds incredible." sounds incredible." I want to try this. I want to try this. I want to try this. The easiest way The easiest way The easiest way to do this is through Google to do this is through Google to do this is through Google AI Studio. We spent AI Studio. We spent AI Studio. We spent a lot of money on the a lot of money on the a lot of money on the ai.studio domain. So I would be ai.studio domain. So I would be ai.studio domain. So I would be grateful if you would grateful if you would grateful if you would use it use it use it . So that we get . So that we get . So that we get some return on some return on some return on investment. So, investment. So, investment. So, ai.studio/live. You can ai.studio/live. You can ai.studio/live. You can try out the model. try out the model. try out the model. You can You can You can chat with her chat with her chat with her and understand the available and understand the available and understand the available options. Let's options. Let's options. Let's do it now. We'll do it now. We'll do it now. We'll just move to just move to just move to ai.dev. We also have ai.dev. ai.dev. We also have ai.dev. ai.dev. We also have ai.dev. Please Please Please use it use it use it too. It wasn't cheap either too. It wasn't cheap either too. It wasn't cheap either . That's it. . That's it. . That's it. ai.dev/live. Here we ai.dev/live. Here we ai.dev/live. Here we can also try out a can also try out a can also try out a live live live translation model, if translation model, if translation model, if you want. But here we you want. But here we you want. But here we can choose our can choose our can choose our Gemini 3.1 Flash Live model. Gemini 3.1 Flash Live model. Gemini 3.1 Flash Live model. We can choose a We can choose a We can choose a different voice. different voice. different voice. For example, I For example, I For example, I like Pac. like Pac. like Pac. Additionally, we can Additionally, we can Additionally, we can enable enable enable justification justification justification using Google Search. using Google Search. using Google Search. Obviously, the model Obviously, the model Obviously, the model is trained on a large is trained on a large is trained on a large amount of information, amount of information, amount of information, but it doesn't have but it doesn't have but it doesn't have access to access to access to real-time data, real-time data, real-time data, right? So, by giving her right? So, by giving her right? So, by giving her access to Google Search, access to Google Search, access to Google Search, she can search for she can search for she can search for information on her information on her information on her own. This is own. This is own. This is essentially like a built-in essentially like a built-in essentially like a built-in function call. And function call. And function call. And if we ask her if we ask her if we ask her about today's about today's about today's weather or, say, weather or, say, weather or, say, why Germany why Germany why Germany played so well at the played so well at the played so well at the World Cup, we World Cup, we World Cup, we can do that. And can do that. And can do that. And she can she can she can use Google to use Google to use Google to find the answer.

find the answer. find the answer. Okay, let's Okay, let's Okay, let's try. Hello, can you try. Hello, can you try. Hello, can you hear me? hear me? Yes, I can hear you perfectly. Yes, I can hear you perfectly. How can I How can I How can I help you today? Do help you today? Do help you today? Do you see me? you see me? you see me? Yes, I see you clearly. Yes, I see you clearly. Yes, I see you clearly. You are wearing a black shirt You are wearing a black shirt You are wearing a black shirt and a black hat, and a black hat, and a black hat, sitting on a blue sitting on a blue sitting on a blue background. Do you want to show background. Do you want to show background. Do you want to show me something specific? me something specific? me something specific? So, what does it say on So, what does it say on So, what does it say on my sweatshirt? my sweatshirt? my sweatshirt? I see the word "Gemini" I see the word "Gemini" I see the word "Gemini" printed in yellow printed in yellow printed in yellow letters. Do you want letters. Do you want letters. Do you want me to me to me to identify anything else? identify anything else? identify anything else? So, what does this mean So, what does this mean So, what does this mean in German? in German? in German? In German it means In German it means In German it means "twins." Would you like "twins." Would you like "twins." Would you like to translate something else? to translate something else? This means I see This means I see you, grandma. Oh, I you, grandma. Oh, I you, grandma. Oh, I see you. You see you. You see you. You look great, grandma. Do you want look great, grandma. Do you want look great, grandma. Do you want to translate something else? to translate something else? Oh, okay, I guess Oh, okay, I guess my Chinese didn't turn out my Chinese didn't turn out my Chinese didn't turn out too well. too well. too well. Can you also Can you also Can you also speak speak speak Chinese, Chinese, Chinese, please? please? please? Yes, I can Yes, I can Yes, I can speak speak speak Chinese too. What would you Chinese too. What would you Chinese too. What would you like to say now? Did like to say now? Did like to say now? Did you say something? Have you say something? Have you say something? Have you eaten yet?

you eaten yet? I don't need to eat because I'm an artificial because I'm an artificial because I'm an artificial intelligence. And you? Have you intelligence. And you? Have you intelligence. And you? Have you eaten yet? eaten yet? eaten yet? I've already eaten, thank you. I've already eaten, thank you. What delicious thing did you eat ? Can you share with ? Can you share with ? Can you share with me? me? Okay, if you don't Okay, if you don't read Chinese, I read Chinese, I read Chinese, I apologize. I was just apologize. I was just apologize. I was just asking the model what asking the model what asking the model what she ate. And she just she ate. And she just she ate. And she just said that she doesn't said that she doesn't said that she doesn't need to eat because need to eat because need to eat because she's a model. So, she's a model. So, she's a model. So, somehow. That's all. somehow. That's all. somehow. That's all. OK. So, this is the OK. So, this is the OK. So, this is the easiest way to get started easiest way to get started easiest way to get started with Gemini with Gemini with Gemini Life at ai.studio/life. If you Life at ai.studio/life. If you Life at ai.studio/life. If you want to learn want to learn want to learn more about developing with more about developing with more about developing with Gemini and the Gemini Life API, Gemini and the Gemini Life API, Gemini and the Gemini Life API, join me join me join me tomorrow morning in Hall tomorrow morning in Hall tomorrow morning in Hall 2009, I think. 2009, I think. Check the schedule. Check the schedule. It sometimes changes. It sometimes changes. And besides that, And besides that, thank you very much. Thank you. See you later.