What powers the next generation of digital commerce? Two billion people, four continents, one system. 10 trillion tokens of data, 500 million interactions every single day. Data no other company holds. We're building the brain of e-commerce, AI at the core, scaled by design. A trillion-dollar ambition already in motion.
Hello, everyone. Good afternoon. Good afternoon, everyone. Welcome to our house. Welcome to our home. How are you?
Good.
Hello. I think it's working. Welcome, everyone. I'm very glad to welcome you all in our AI house. AI house was opened by Prosus like just three months ago, and we had 40-45 events just in three months. Every week we have hundreds of people here thinking about the future of AI in Europe, in the world, sharing, showing their ideas, their startups, their investments. We are super proud to have this place to be the meeting point of AI in Amsterdam. Today we are happy to receive you here in our house, but we are going to talk not about the future of AI in Netherlands, but the future of Prosus. Very happy to have you all here, and I think it's going to be a very fun afternoon and event.
Today we're not going to talk about finance or goals or numbers. We are going to talk to open the curtains. Prosus is working a lot on AI innovation. Prosus is working in building the next generation of services, and we are going to talk to you about our vision. This is a tech event. I'm just going to do the intro, then I leave, and we are going to see demos and products and live demonstrations and versions of products that we haven't launched yet. I think it's going to be very fun. I forgot to say, we have lots of visitors, partners and Prosus team here, but we also have a webcast, so good morning, good afternoon, good evening, everyone that is watching us remote, on the webcast, and hope you also enjoy today.
We are going to talk a lot about our vision, our future, and how we are going to together build an AI future. We are going to also release lots of tools today, open source tools, platforms, that we are going to work not only inside the whole group, but with our partners, but with the AI community. Lots of fun today. Let's go. Let's go. Our agenda today, I want to start remind you the Prosus that you already know. This is the Prosus that you know. Just to remember a little about us. We have more or less one billion customers, more or less five million partners. We transact around $165 billion in sales.
The more important thing is that we are running a lifetime ecosystem where we deliver high frequency service to our customers, like food delivery or quick commerce, like pharmacy or grocery. High frequency also payments and banking services. Our customers use us very often on these parts, but then it enable us to cross-sell to some high ticket services. For example, when you book a trip or your rides or your events, or even when you buy a new car or a new house. That's what Prosus is doing today. That's how we sell $65 billion, and we are very, very strong on this ecosystem that on this flywheel, specifically in Latin America, where we have great brands like iFood or OLX Brazil, or and in India, where we have like Swiggy, Rapido, PayU, lots of great companies.
Here in Europe, where we have Just Eat, OLX, eMAG. That's the Prosus that you know today. I mean, we are here today to talk a little about the future. What we've been doing over the last few months that usually we never disclose it as much as you are going to see today, and what we think is going to happen over the next few months, then years. First I want to tell you, we have deep AI foundations. Today, AI is fashion. Everyone talks about that. We started doing that in 2018. Actually, I don't know if you know, I was the CEO of iFood by the time, and Prosus invited me in 2018 to travel together with Prosus to start dreaming about the future of technology.
That's how iFood started to invest a lot in AI. After that, iFood, I think it's one of the best AI companies in the world, the whole company today run on top of an AI model. What I want to share with you, when I’m CEO just for one year and a half, what I told you is Prosus is going to be also an AI company and a global class AI company, and that's what I want to share more with you today. Today, over the last two years, AI in Prosus grow a lot. Today we have AI embedded in our culture. We have more than 40,000 agents running with our 40,000 employees, and we are going to see that working today. Hope it inspires you a little.
We are going to talk today about our Prosus AI Labs. We have an AI lab that is working in research, in new technology, in open source software, in new models, training new models. We are going to talk about how we have a unique set of data that enable us to create AI personalization to understand about customers and transactions, specifically in India, Latin America and Europe, as no other company can do, and we are going to show it to you running. We have hundreds of models, and today we are telling you a little more about the next generation of AI models, what we call Large Commerce Model. It's one big model that can substitute hundreds of models in Prosus, and we have thousands of people in technology, at least a thousand in our AI area.
Prosus is already an AI company, but we don't use to talk about that. That's what we are going to show you today. Hope you enjoy as much as me. I have a lot of fun with all of that. Today, the Prosus of today, it has a glue around loyalty. All those products here, we can cross-sell them through loyalty services. For example, we talk a lot about Clube iFood, where we can use the iFood frequency to sell many other products, for example, Despegar or other products, to our customers. We are increasing our loyalty presence to India and Europe too, and that's the foundation of our glue on our ecosystem. The world is changing a lot. The world in one or two years is going to be completely different, and we are working to build this new world.
What we believe is that the interface that we interact with computers changed a lot over the last years. It was first speech, paper, websites. In the last 15 years, apps became the real way to interact with your digital life. We are seeing now a transformation where agents is going to be the most important point of contact between you and your digital services, your digital companion. We really are going to see a big interface change over the next probably one to two years. This is going to change how all companies in the world work, and this is going to change how Prosus work. We are creating this future, and we want to share with you how we are creating this future. How is that? We are going to keep offering our loyalty.
I like very much our loyalty offering, but every time more, the glue comes from loyalty to life assistance and agentic OS. I want to introduce to you these two concepts now to start. Life assistance for us is not like an assistant the way that you have one. You have ChatGPT and Gemini or Grok, and you can ask him, them, "Tell me about that. Tell me how to prepare this food," or, "Tell me about the history of this country." So it's very good to help you to write something or answer some question. But we are going to see this year a quick change from assistant that answer information to assistant that do things. That's what Prosus is focused on. You are going to say to your assistant, "Do that for me. Give me food in one hour.
Buy this ticket to me. Pay this bill for me," and he's going to handle. "Schedule my haircut close to where I am." Everything is going to work and pay, and it's going to be much simpler than how you do that. We want to share with you a little about that. To have life assistance, we need to change also how the companies work. The entrepreneurs that run the company, they need to have tools that also automates how the company works. The companies every time more, we have agents working on that. We believe more than that. The companies will work automatically throughout and become autonomous agents. I'm going to talk one minute about each of them, and then I stop talking because today's a tech day, so we are going to demonstrate everything.
First, on life assistant. This is iFood app. Everyone loves it. We have 70 million customers per year using and buying food there. The best users, they spend five minutes to order in average food in iFood. This is an amazing app that everyone loves. They take five minutes to scroll, select, think how they do something. What we are talking about is, can we do that in 1 minute? That's our goal. You just take your phone and just say, "I want Japanese at 1," and they find out where you are, what you like, how much you want to pay, where, what is your payment method, what kind of voucher you need, and deliver that to you and keep track about it's arriving, it's late, it's not arriving. If it is Just Eat and iFood, it is arriving on time.
If it is the competition, no, it's late. The assistant is going to solve all those problems, and we are going to show some of that to you today. This is a trend, but the food delivery that I'm talking about is just one case, the fast order like that. We have 100 case that we are designing, developing, test with real users, testing frequency and retention, and understanding how is the future of interface, and we are going to show maybe five of these cases today. We are working with 100, and I really think we are going to push how people interact with their services. To have a AI assistance, we also need companies that are really working with Artificial Intelligence , really work with intelligence.
We are going to show you today that first process really work with intelligence. We have Toqan for Work, how we call it, and we have thousands, tens of thousands of agents that makes the people here work faster and better. That's great. What we really want to show you today. We want to show both. After showing that, we want to show you Toqan for Partners, and that's a big difference we are starting now. We developed a big platform, the Toqan platform, and we are going to open that for all our five million partners. We start now with a few of them, but our intention is keep opening, and opening, and provide for restaurants or hotels or dealers Artificial Intelligence to them. Not only make questions, but also marketing agents, promotion agents, finance agents, planning, consulting services.
If you think nice PowerPoint, wait a little more because we are going to show it for you in a few minutes. Before I move on, I want to say, what I just described to you are going to see now, but this is just the start. We really believe in agentic companies. Today, you can have an agent or a manager here that do accounting, asking an agent to fix some problem, but that's not the future. That's the reality today. You're going to see it running today. The future is the whole company's autonomous. This is the entrepreneur, the founder, the CEO of the company, and he will have an autonomous operating system that is going to say, "This is our website. Let's change how we write ourselves there. This is our financial numbers.
I'm worried we might miss cash in 20-30 days. I suggest we invest more in marketing and less in some other area." We believe we will have soon agents that run the whole organization, and they can offer options to the founder, to the entrepreneur, and he can decide what he want to execute or not, and he can ask, "Please execute this thing." You are going to see today the foundations of that, but that's the vision we are implementing, and that's the vision we expect to release starting in restaurants over the next few, I would say weeks, because I'm a little, like, psychotic to move faster, but let's say month. That's what we are working on, and you are going to see the foundation of that today. That's our agenda for today. It's a tech event.
We will have tech people talking here. I'm feeling bad now. I'm also tech people. The tech people that are really developing, they are going to be here, not me. We are going to start talking about Large Commerce Model, and Large Commerce Model is the foundation of everything we are doing. To train the language models on the internet, people train them using real-world data and all the books and everything on the internet. We are training our model with all the 500 million transactions we have per day. We can learn a lot about customers, and we want to show you why this is important. It is very important to iFood today, and now it's moving to all other companies. Second, we'll talk about the future of work. First, Toqan for Work and how everyone in Prosus works faster today.
Toqan for Partners, what we are pushing to our partners all around. Then we finish with the future of apps. I don't like this name very much, but the future of the way you interact with your computer, your mobile phone, your cloud, and we think it's going to be different. We have lots of systems running. With that, we are going to demonstrate to you lots of things. I want to remind you, what you are going to see today, there is one or two things that are available just for 1,000 or 5,000 people, but many of them are running in Brazil to millions of users through iFood, in Argentina to millions of users through Despegar, in Poland through OLX, in Germany through Just Eat, in Bangalore.
You are going to see things that millions and millions of people are already using, but we will make it, we will deliver it to billions of people, and we want to push from Europe, Brazil, Latin America and India to the world. We think very good innovation and help to create how the world keep evolving and offering better services and offering better support tools for our partners. I'm very excited about that. Hope you enjoy. The maestro of the event today is going to be Euro, our Head of AI. Let's move on. Video and Euro coming here to teach us a little more about the future of work. Thank you, and see you soon.
A Tuesday craving. A 3:00 A.M. search you forgot by morning. The hotel you looked at twice but never booked. The thing in your basket you couldn't quite justify. You scroll, you pause, you abandon, you come back, and somewhere in all of that chaos, a pattern begins to form. Not who you said you were, who you actually are. The person who runs at dawn and eats ramen in the rain. Who buys a pink fedora hat and never explains why. You are not defined by one single data point. Billions of data. Unique human moments.
Welcome, everyone. Good to see you here. As you can see from the clip, if we are into pottery, we have a system for that. My role today is to showcase the great work that all our colleagues across the group are doing and the technologies, some of those that are already deployed at scale, used by millions of people, but also those that we're working now, we are going to release in the next months, but also some that we're just experimenting with. They're super cool technologies, might take a while, and we want also your input, and we'd really like you to test them as soon as we can. First, I want to start with Large Commerce Models. As Fabrício said, that's a foundation of many of the things that we do. The foundation is a very specific reason.
500 million customer interactions every day that produce
Billions of data points. We have been using this data for years. A lot of the recommendation that you see in our system, recommendations that power already our systems are really based on this data. Fraud detection, all these things, you are familiar with that. That's really the case. However, we never had technologies that really understand this data. We never could reason about this data in the way that we humans do. That's the missing point. Now, this is something which, in fact, come on to everything about AI. There is something interesting here, that the most valuable AI won't be trained on text. We come to that realization after having used large language models for years, we come to that point. The real intelligence is trained on real-world interactions, and we have that. That's our combination.
We know how to do AI, we have the real world interactions, and that's what we do. We call them Large Commerce Models. In fact, they are large language models trained in a very specific way to be good at commerce. We train them on billions of interactions instead of billion of documents. They learn intent and preferences instead of learning language. Of course, once you have that, you can recommend, personalize, forecast instead of translating, summarizing, and doing the things that we do with large language models. Large Commerce Models is what powers the applications, both for our own consumers, but also for our own partners. They of course stay behind everything that we do. They are really, let's say the foundation of everything that we do. I'm really keen to show you how that works.
For that one, I need help. Zülküf, please come on stage and show us how LCM work.
Thank you, Yaro. Hi, everyone. I'm really excited to be here today and show LCM, Large Commerce Model, in action. LCM was a concept a year ago, but now it's in the production at scale, and we are using LCM intelligence across over 40 applications in iFood. Today, I picked two to show you, one from the consumer side and one from the business side. Let's start with the consumer side to see this LCM's deep user understanding, what it means actually in practice. Imagine, we all are iFood users. We are in Brazil, we take our phone, we open the iFood app, and most probably none of us would see the same home screen there. It would be all personalized for each of us. To make this more clear, I picked two users, two real iFood users from both ends.
On the right, we have Lucas. He's a premium explorer. He comes to app, and he likes to explore. That's why on top, we have a cuisine grid to explore by taste. Below that, we have premium restaurants targeting him and curated content based on his taste profile. What we don't see here is promotions, vouchers, nothing about the price, but everything about the quality. On the left side, we have Dimas. It's a different story. He's a deal seeker, which means he only buys with a good deal. That's why right on top, he sees a promotional banner, and price is in the focus point of everything. You see all the prices are crossed out. Discount vouchers are everywhere. For him, that's what matters.
I want to pause here for a minute because you may think, "Okay, that's a recommendation." That's not a typical recommendation engine ranking the offers differently for each user. That's both homepages are entirely generated by LCM, our Large Commerce Model. The content, the layout, the visuals, the entire experience has been generated for these users specifically, and that was not possible before. How does it work? I want to take you behind the scenes and show you how LCM works without giving you the secret sauce. This is LCM thinking. LCM thinking is our LCM is a reasoning model, and thinking here is connecting the dots. It takes very rich user insights and turn them to the action points, and in that case, turning them to the home screen. How does it do it?
It can do it because we have trained that model on billions of iFood transactions, and it learn what is the ultimate user experience, what is the best outcome, and how to use these insights to create that one. That model learns how to use this data. We cannot read it, but I can highlight a couple of points here. For example, the reason we didn't see any promotion on the Lucas screen, because he has zero sensitivity to the promotions. He doesn't care. He didn't even filter once the list by price, because for him, the only thing that matters is the quality. He goes, rates the reviews, and buy from the top restaurants.
That's why LCM even remove that promotion carousel that's always there and replace it with more premium content, even pick the color palette that give this, you know, exclusive feeling for him. On the other side, Dimas is a different story because the only thing he cares is the price. He didn't place any order without promotions. He abandoned his cart when he sees a small delivery fee. For him, price is the most important decision point. That's why we see all this promotional content around. Single app, two very different experiences created by LCM. It doesn't have to be screen, huh? It can be WhatsApp conversation, it can be voice interaction. The surface can change, but the underlying intelligence, the deep understanding of user is always there by LCM. This is the consumer side. I'm going to take you to business side.
In iFood, we have many business teams every day working on their goals, like growing a category, reducing the churn users, activating some users. They have many goals. In these goals, promotion campaigns is one of the most important leverages like to reach that goals, and yet they are one of the most expensive ones. It really matters to get these promotions right. What that means is to know who to target and to know how to approach that target. Otherwise, if you don't know who to target and how to approach them, you can easily waste your promotion budget and you don't reach your goal. That's very critical. How it is done in the industry today, there are big teams. The data teams, they come up with hypothesis to improve this, reach that goals.
They go analyze the data, come up with results, train the models. They spend months to really organize good promotion campaigns. Now I'm going to show you how we can do that with LCM. Imagine I'm getting in the shoes of iFood business owner, and I go look at my dashboard and I see oh my god, in my region, dessert category is underperforming. I want to sell more desserts. What I do, I take LCM Studio. Actually, this is just a tool interface to the LCM intelligence. Normally we do that over APIs, but just for demonstration, I'm using this tool and asking the question. LCM starts thinking, "Okay, I have a goal and I want to brainstorm and get some ideas before actually designing that promotion campaign." It comes up with couple of ideas.
The first idea is people who browse but they don't buy. There are users. Remember, our goal is to increase the sales of the desserts. We want to sell more to people who are potential buyers, but they don't buy. The first idea, we can target people who already browse, but for some reason, they don't order it. Some people, they already buy sweet stuff like fruit juice, but they don't place any dessert order. People who already buy a lot of stuff, but for some reason they don't buy dessert. Those are ideas. Already good start, but those are just ideas. We don't know if there is a volume behind it, like if there is a size. Do we really have these users in our region or that's just a model hallucination, right?
We need to see these ideas have actual representation in our user base. How is this done traditionally is data team comes with the business analyst, they take those ideas, they go to data, they write lots of SQL queries, analyze it, sometimes train the models because there is no a flag indicating this person is a potential dessert buyer. They need to predict actually who can buy. Again, weeks of work, if not months. Here I can just say size all those three. What it does, it actually takes those ideas, and it goes to our user base and try to see actually can we fill in these categories. Do we have enough users represented for these categories? That takes couple of minutes. I ran it before, so I will cheat here a little and show you here.
I ran it before, and when we look at there, what I see, the most promising category is people who browse the dessert but they don't buy. It is like 29.9% of people in that neighborhood, at least once in last three months browsed the dessert category, but they didn't buy. We can see actual data about them, how many orders they do, their conversion rate and everything. This is like in minutes, I already at that point, I saved months of work actually. That's not enough because now I know my target group, but I don't know how to approach them, right? These people are there, but what should I do to sell more dessert to them? Then here it, where it gets interesting, I can also survey. Run a survey with this group and ask them, "Why don't you buy?
I mean, what blocks you buying a dessert? Normally, again, surveying means we take this user group and a team calls these people, and then it takes weeks of time. Now, I don't need to run actual survey. What I can do, LCM knows these people very well, the user group. Then we can simulate the survey. It's a synthetic survey. We can ask, and LCM gets in the shoes of these users and can answer these questions. Why that specific person doesn't buy a dessert? Because he doesn't like, because delivery fee is so high, because he doesn't like the merchant, et cetera. There can be many reasons. With all these trainings on the billions of transactions, LCM can come to point that can give very informed predictions about that. Actually, none of the surveys gives us 100% accuracy.
Here, what we are looking for is not a precision, but a direction where we should go and how to approach these people. Based on the results, what we see, high delivery fee is the biggest blockage for these people. If we are going to organize a campaign, we better incentivize high delivery fee. We give a promotion on the delivery fee, then we can have the highest chance of success. Actually, this is still, I'm in the loop and as a business analyst, I run that. What we are working on now, we want to also give that to agent. Like, I want to come up with this goal just before leaving the office, give it to my Toqan agent, Sean will talk about it, and then I leave the office.
When I'm at home, the agent keeps working, comes up with the ideas, runs surveys, runs campaigns, and next day when I come back, I see all the strategies, how they work, and I pick one and execute it. We did actually something very similar without the agent and achieve very good results. I invite Euro come in here and then explain those results. Thank you very much.
Thank you. Thank you very much. We started working on LCM about a year and a half ago. There was a lot of science that we have to create because LLMs do not really do what we need them to do and so on. Now we are at the point in which companies like iFood already benefit significantly from this. The other things that we see is that the limiting factor is our imagination at this point, right? Because you can see you have all this freedom to invent different service and different use cases. The great team at iFood, as I said, has been the first one to deploy this at scale. They deployed it in many, many use cases, about 40. The homepage, so let's say the place where you start, you can see the impact, the business impact.
Also everything that has to do with advertisement. That's very important because you can cut the advertisement budget significantly for the same effect. Also notification. Notification is the messages that you receive, and they are spam or they can be great, right? Let me show you how that can be great. Anyway, at this moment, LCM powers more or less 15 million orders, so it's a significant amount, 40 use cases. I want to underline this. We make this model really, really inexpensive compared to anything out there. That's a condition to run this at scale. Here we are running trillions of transactions. These models have to be inexpensive for the return on investment to be positive. We have done that. Sometimes we are surprised. This is a user that posted that comment on social. This is the notification.
Notification can be spam or can be delightful, right? "Hey, was it down? I missed my push today." That person was complaining they didn't receive the push, right? Why is that? Because it's relevant, because it's funny, because it's useful, right? You change spam into something which is a delightful experience, and I think that's pretty important. LCM is now rolling out across the entire group. We started with iFood, but this goes everywhere. Just Eat, OLX, eMAG, and many others just started, but you can see already the results and we are super excited about the whole importance that LCM will have for our services and products. Let me change gear now because I want to talk about agents at work.
A few years back, around two years ago, when we were deep into experimentation with large language models, some of our colleagues were coming back with use cases that we didn't expect. They were use cases that really accelerated their way of working. We decided to facilitate that. The way that we decided to facilitate that is with giving people tools where they can create their own agents and they can automate their own work. There is an idea behind, an idea that we really believe in, that we started here with chatbots, which just give, let's say, interesting responses, but no more than that. We started using tools that automate jobs, which is more or less where we are now. They do the entire work, entire workflow, something that might take 30 minutes, 45 minutes, one hour.
Eventually that will be the point where jobs change, and after that, organizations change. There is a logic in what we do here. We see that that logic enabling all this will make us more productive, more interesting, and can do better. To do that, we created a platform which is called Toqan. The idea of Toqan is that every employee across Prosus can create agents for themselves. Everybody's a builder. There are no developers and everybody else. Everybody's a builder. There are about 25,000 associates that use these tools at this moment. To show them how they work, I need, again, some help. Sean.
Thank you so much, Euro. Hi, everybody. I'm Sean, Principal Product Manager here at Prosus, and I'm responsible for Toqan, our agentic workforce platform. I'm going to take a few minutes to explain what's under the hood, and I'll do that by showing you how users can use Toqan to build custom agents that help them solve key operational challenges in their workflows. I'm going to do that by picking one concrete example. In this case, we'll start looking at account managers in our food delivery platforms. These account managers are responsible for the supply side of the marketplaces. They make sure that the restaurants are happy with the partnership, that they're performing well, and they're growing together with us. They also make sure that the restaurants are continuing to spend more of their time, energy, and budget on our apps and services.
It's a very difficult job, because in order to help restaurants grow, you have to navigate an incredibly complex data landscape. You have to be able to pinpoint and find tiny details in the data, connect them with the story about the restaurant in order to help them grow by giving them the right advice. Today, these account managers do that by preparing for weekly and monthly updates by digging through data for hours, compiling reports and trying to understand how exactly they can help the restaurants when they go and visit them. That means that we both are sometimes missing key data points when we're helping the restaurants, and we're not able to cover all of the restaurants in our portfolio because that would require way too much time from these account managers.
Now, luckily, these account managers have built a solution in Toqan themselves, which we're going to showcase in this case. The account managers in Toqan. This is Toqan, by the way. This is a platform that supports about 25,000 users on a monthly basis. In this platform, the account managers went into the Agent tab, which is where they can configure and build an agent that solves their unique operational problem. Let's go and take a look at what they've built. We'll open the Agents tab, and we'll talk a bit more about what you see. We immediately see the account manager agent, and we'll just take a look into how they've set that up. Let's go and edit that. Needless to say, we're not showing sensitive iFood information today, so we've masked the data.
What you'll see will look Dutch, but this is based on exactly what the setup is that iFood has done. You can see here, the first thing you see is a big Instructions tab. What the account managers have done here is simply explain what the problem is and explain to Toqan how they should solve it. In a way, natural language has become the programming language of moving the organization forward. It's incredibly cool because not only does that mean that our Prosus users are power users of this technology, but everybody becomes a builder that fundamentally can change how their organizations do work. With this, the agent knows how it should work and what it needs to solve. We need to make sure it actually has the ability to do that. To do so, these account managers give Toqan tools.
Tools are the ability to integrate and interact with systems, taking real actions and not just providing an answer. In this case, you can see that the account managers have connected Toqan to Databricks, which is where iFood stores all of their operational data around restaurants and many more things. Anything you want to know about restaurants, you'll find it here. The difficulty beforehand was that for account managers, this platform is not easy to use. It requires deep technical expertise, and it's difficult to manage access, and especially if you're on the road, it's not always easy to navigate going to your next meeting and being able to dig through the data here. For Toqan, that's no problem. The only other thing that Toqan needs is a Salesforce connection, which is where we maintain our relationships with the restaurants.
Here we find what we've done with the restaurants before. We can track key events, and we generally can understand the account that we're dealing with. With all of this being done, the account managers essentially have taught Toqan exactly what it needs to know in order to do the work. Now let's put ourselves into the shoes of one of these account managers. Say it's a Monday morning, you're starting work, you have a big portfolio. How do you start? They used to start by digging through emails, their data, and figuring out where and how they should spend attention. A good part of Monday morning was spent on that. Now, we can simply go into the agent, and we can ask a simple question. In this case, I'm going to say, "Hey, Toqan, review my portfolio." Right?
I need to check what's going on. I'm responsible for a large portfolio. How can I help? Identify critical accounts, right? Help me not just do the weekly updates, but I need to know who I need to pay attention on and how can I leverage that best. I ask for what are the problems that these accounts are facing right now, why are they considered critical, and how can I help them, right? How can I improve? How can I make sure that their operations are getting on track again? Let's fire that off.
As you can see, Toqan immediately starts working, and in just a second, it will start reading information from my portfolio on Salesforce, and it'll continue to go through many, many tasks, compiling data, kind of digging gradually through where it needs to pay attention, what it needs to do and needs to know in order to help me do my work. Now, this takes a few minutes to run because it's a very complex task, many steps involved, which sounds long, but in actuality, these account managers had to spend hours on each of these deep dives beforehand. But for the sake of your time, I'm going to skip ahead. Just like Zülküf, I'll cheat a little bit, and I'll go to a conversation that we ran just before with exactly the same question. Here you can see how Toqan responded.
The first thing you'll see, again, Toqan makes this very readable, very user-friendly. It tells me immediately, "Hey, I'm looking at your portfolio. There's 5 accounts I'm seeing specifically, and there's two that immediately you need to pay attention to." Also here, you can see the Dutch version of many of the iFood cuisines. You can see Toqan highlights these as critical, and there we see a rapidly decreasing trend, so we better get on that. Toqan doesn't just give me an overview over my portfolio. It starts giving me an in-depth breakdown for both of these accounts. Here we're looking at the first account. You see kind of some information in general over the account, and then we get what we call the story.
Here, Toqan starts to combine data points both from the operational data and the data that we know about the account from Salesforce and starts painting a picture, not only why and where are we seeing performance decreases. Here, for example, we find a massive decrease where we see that we almost go from 1%- 1.2% of cancellation rates to almost 8%, but Toqan already is able to connect that to a change in operations lead several months back. Since then, we've seen a decrease in performance. Now, Toqan highlights a lot of things here, but just to skim through this a bit more, Toqan doesn't just highlight some of these story points, it also shows you what is the root cause, right? What are some of the elements that may have triggered some of these issues?
It doesn't just show the root cause. Like, for example, there's 10 locations in particular that are failing, but it also highlights what can we do about it, right? How could the account manager now go and help this restaurant start fixing it? Maybe we should visit some of these restaurants. Maybe we should sit together with the new operations director to see how we can help them get their organization back on track. You can imagine from this depth of information, and this is only the first of the two accounts that we get a breakdown from, how long this would have taken an account manager to prepare. Now, with a simple ask of a question, in five minutes in which they can check their emails, they get a full breakdown of the account, what's going wrong, and how they can help them.
This is only one question. This agent itself is used daily by over 800 account executives at iFood, answering over 40,000 requests every single month. That means that we're providing these restaurants not just with better and faster support, but we're also able to cover many, many more restaurants in our ecosystem, giving them a better experience on our platforms. One benefit that comes from building this as a process ecosystem is that this agent isn't just isolated to iFood, but actually it's an agent that we can go and replicate and share with other portfolio companies. This specific agent has already been shared within OLX, with Just Eat, with Despegar, and we're rolling it out to more portfolio companies as we're speaking. Toqan is also not just a platform for one agent, but it's a platform where anybody can build agents that are operationally helpful.
5,000 agents are active every single day across many different departments and domains of our businesses. You can, for example, see that we have agents in procurement that help us ingest contract data and refine that so we can get better insights and negotiate better deals. You see that we update our invoice mapping, which helps us reduce a huge amount of manual work while getting better tax claims when we file for taxes. There's many more agents just like this, shared and built by individuals themselves across the portfolio. Now, what you've seen now was an agent that I as a user have to go and have to talk to, and together we can start solving complex problems together that make our companies work better. I still have to chat to it.
Recently we've made a big release where we introduced Toqan background tasks. The concept of agents working in a background doing work before you even know work had to be done. I would love to introduce that feature to you the same way we introduced it to our users with a short demo video.
Toqan already connects to your entire stack, GitHub, Jira, Confluence, Google Workspace, and more. You can even build custom tools for your internal systems. With MCP support, the possibilities are endless. There's one problem. Toqan only works when you ask it to. Toqan waits. It has access. It knows how to help, but it can't start until you ask. Introducing Tasks, agents doing the work before you even know there's work to do. Set up a trigger, a webhook for any system or a schedule, and choose an agent to pick it up. Easily define what an agent should do, triggered by adjusting the system front. Now the agent starts working the moment something happens. New lead? An agent researches the company, finds case studies, drafts outreach instantly. Daily standup reports, weekly analytics, monthly summaries, all automatic. Multiple Toqan agents working 24/7, triggered by real world events.
That's not just an agent, that's an AI workforce.
Brilliant. This was super interesting to us, and our users were super excited, and I'm sure we're not the only ones who can see how this can fundamentally change how work gets done in and across companies. With that being said, I think there's a mandatory closing line, which is we're just getting started. With that, I'd love to hand it back to Euro Beinat to finish off the session.
Sean, thank you very much. I think you should agree, or you agree with me, Sean is great at making complex things simple, right? He's just as good. The thing is that this has become simple, right? It's not only Sean, it is that all these tools now are accessible for everybody. You don't need engineers for doing this. If you can write, you can benefit from agents. It's a really important democratization of everything. We have mentioned a couple of times about 40,000 agents being created across processes. Why do we focus about such a large number? I mean, who cares? It's just a big number. We care because it is a cultural shift. It means that people now can do it. There's no barrier. Everybody is brought into this. That's what is the effect of the 40,000. 40,000 means different culture.
What matters, 5,000 that are used every day. These 5,000 are used every day because they have value. 40,000 changes the culture, 5,000 changes the company, right? They're used everywhere, in operations, in HR, in legal, in sales, in marketing, and of course in engineering. Everywhere. Everybody becomes a builder. That's how an organization change. If you want to really ride the benefits of this, organizations have to be different. There's also another thing that we have seen over time. It really pleases us. We started deploying agents and everybody started building, say a bit over a year ago. Initially, most of them were simple, the black ones. Now see what happens over time. The red ones are the complicated ones. They tap around different systems. They stitch together very complicated data. These are the agents that have value.
The proportions of the agents, complex agents, grows all the time. It means that people create agents that help them more. These take about 750,000 actions every month. We estimate that they do the work about 1,000 people at least. Now, this is a notoriously difficult thing to measure, but what's sort of more important, what people say, "We have more agility, more independence. We can do more things independent of the time saving." This is how organization change. We're really proud about all this work, and we continue working on this. Once we have the LCM, once we have this experience of deploying agents in our organization, why don't we offer all that to our partners? Why don't we get our partners to benefit about that too? That's exactly what we want to show you next.
Behind every small business is someone doing three jobs at once. What if they didn't have to?
Hey, Fernando, it's Tiago from iFood Pago. Impressive growth lately. You've been pre-approved for up to $12,000 in credit. What are the monthly payments like? I just ran a simulation for you. Here's a breakdown of the best payment conditions for your cash flow.
No bank branches, no waiting rooms, no chasing documents, just a partner who knows your business.
I've got 200 listings. Which ones need attention? Your top five listings are underperforming. Here's what to do. Done.
AutoIQ, smart summaries, one-click actions built for car dealers who want results now. This is what a partner feels like. Intelligence that has your back every hour of the day.
Here is that the best time to ask for credit is while you cook. Huh? So keep that in mind next time you're in the kitchen. That's a good time to ask credit. What do we mean with agent for partners? We are looking at the technology that we already use inside LCM and agent that are so impactful for us, and we look at ways to offer all this to our partners, restaurants, dealers, hotels, all the partners in our portfolio everywhere in Prosus. How? What does that mean? It means powering them, empowering them to use our platform better, to start selling on our platforms better, but to grow their business, to make their operations better, excel in their operations. Finally, as Fabrizio showed at the beginning, maybe to automate part of their work. They might be good at cooking, but not necessarily at managing, right?
There's a different thing. We already started doing this, and there are already hundreds of these agents working. Going to show you one example, and then I'm going to ask for help again to show specifically how this works. This is an agent which helps onboard partners, in this case, restaurants on our platform. Here you can see what happens over time. First of all, the number of restaurants that are onboarded this year. But there are two interesting effects here of using these agents. 80% of this onboarding is aided by agents, and about 20% is completely automatic. You don't have anybody on the other side. It's all taken care of by the agent. 20% of that is completely automatic. At the same time, 60% less sales reps onboard 18% more partners. We change, our partners change.
These things affect us both in a very deep way. To show you more in specific in detail how that works, the great Ioannis.
Thank you, Euro. Let me plug in. Hi, welcome everyone. My name is Ioannis. I'm leading Tocan, and I want to show you how the same technology that we built for our agents at work, we are now deploying for our partners. Also, hi to those online. Let's go with a demo, and I want to start with Tiago. Tiago, he lives in São Paulo, and he's really, really good at making açaí. I hope I'm pronouncing it correct. There are lots of Brazilians here. Tiago, açaí is a frozen yogurt-ish, fruity delicatessen, and he's really good at it. He knows that. But he has a problem. He doesn't sell as much as he wants. He wants to know how to increase his business. Well, now he has Tocan. What he can do with Tocan?
First of all, as you can see, we keep the same technology, and we have changed the UI to make it much more user-friendly for Tiago. He can go there, and then he can see a suite of agents which have been built by iFood and also have been augmented by our ecosystem. He can go and he can see agents that he can help him identify which of the locations are actually those that are performing better. Or he can make changes on the menu. Or in the case of Tiago, because he's very good at, you know, cooking but not so good at selling, how to grow my business. He goes into the growth agent. We call our growth agent Cris. We like giving it a name, and our growth agent is actually powered by LCM.
The moment that Tiago goes to the agent and asks the questions, the whole data come together. One of the questions might be, for example, I would like to know. This is a demo account in the sense that I've changed the data behind because you will see real revenue and real numbers. I have altered that to make sure that it's private, but it's all connected. How to increase my revenues and like my neighbors. Neighbors. All right. Great. It seems like a prompt that he can think of. By the way, Cris, at the moment, it's already been live in hundreds of thousands of restaurants in Brazil through iFood, and it lives on WhatsApp. Now what we are doing is that we are centralizing all this powered by Toqan and LCM, and all agents are available for use.
Now what we are seeing is that, okay, he needs to understand. He says, "I need to understand how to increase the revenues compared to a specific location." He goes, he checks the competitor context, which he can only do because he has access to the web, but also he has access to specific data. He can know exactly what are the critical merchant context files, which are coming from the insights that they are coming from LCM. He goes, and he says, "Okay, you know what? I know exactly what you need to do. First, fix your availability." Sometimes the hardest is the simplest solution, right? Fix your availability. Be available. Second, create combos. Great idea. Why you just don't bundle things? Because in that way maybe you can have a lower per order fee. And third, reduce promo dependencies.
He creates a projected impact over this. That can only be done because he's connected to our data. Okay, good. I like it. Now I need some help. Maybe he says, "Okay, good stuff. Good job." Let's see. "I want now a menu to check and then automatically update it, update iFood afterwards." All right, good. This agent cannot directly update iFood because it's a different agent that does this type of work. Now I just want to get the HTML. I've already created an HTML before because I don't want to bother you with opening the file, and it literally looks like this. This is actually the menu that the agent has created specialized for Tiago, the açaí owner, and it can literally be his website if you want, right?
You can literally take this, put a publish button, and then that can be. Even print it and hand it over as a menu. Simple. Very, very simple. Simple and beautiful. Okay, great. Now this is our friend from Brazil. Now let's go from Brazil to Poland and then see how the same technology powers within our ecosystem, a different type of company, OLX. In Poland, we have OLX. In OLX, we have one of the brands in Otomoto. Otomoto, we have, it's a dealership. It's a marketplace where dealers can go and sell cars, and then, you know, the Polish people can buy cars from the dealers. One of the things that we have done there is that we have launched the same technology but lives inside the platform of Otomoto.
We call it AutoIQ, and it has been done with an incredible work from our OLX colleagues. Some of them are here today, and it actually has one simple, very simple goal, make the life of a dealer simple. Don't waste time trying to answer the leads, find information, going through the different kind of mosaic of complex data. Just do what you're good at, which is talking to the people to sell your cars. I have a video for this. Let's go and see it.
Let's take a look at the daily life of Tomasz.
Hello, I'm Tomasz. Nice to meet you.
He runs a family car dealership in Warsaw. The first thing he used to do was log into his Otomoto dealer account and spend 30 minutes assessing his top 10 underperforming ads to decide on actions like reducing the price or applying a visibility boost. Now with the AutoIQ agent, one click gets him the same outcome. He receives a summary of his overall stock performance in under 30 seconds instead of 30 minutes, along with actions to move the stock faster. Previously, if Tomasz had a question about his stock, he used to call his Otomoto account manager or wait for a visit, but now he can simply ask the AutoIQ agent. This helps him make quicker decisions on his inventory so he can sell faster and smarter. Tomasz usually has a couple of cars he needs to post on Otomoto weekly.
Previously, it took him almost 15 minutes to post a single ad, but now he uses AutoIQ from his phone to post at lightning speed. He simply takes a video of the car outside and inside, and the AI extracts the best quality pictures and pre-fills the ad description and details. The only action Tomasz takes is reviewing the details and confirming. Snap, upload, and done. OLX AutoIQ is profoundly impacting the daily lives of car dealerships. By automating routine administrative tasks in their workflow, AutoIQ enables dealers to dedicate more time to interacting with customers on the floor, ultimately leading to faster sales. Our ambitious vision is to create a fully autonomous end-to-end AutoIQ multi-agent capable of managing the entire process from initial car sourcing through to post-sale activities, and we are just starting.
I like Tomasz being happy, and he's just starting, right? That's pretty cool. Come on, Euro. Thank you, Ioannis. There's one thing that really resonates with me is this, all these examples are great, but if I go back to the car example, the dealer example here, it's easy to forget that to be able to do all this automation that you have seen here, which really make life easier for these dealers, you really need to know dealerships. You really need to do that business to know that business inside out, and that's the reason why we can do it, because we know that type of business. We have been working with dealers for years. We know restaurant for years. We know hotel for years. That's the value that we bring in, is not only the technology.
It's a very long history of deep relationship with these partners that make all this possible. Don't forget that because not only technology. We have been in this business for years. That's why we can use the technology in the way that we do. There are several of these highly skilled, very specialized agents already in operation at scale across the group that help thousands of partners. You see them almost everywhere at this moment. They help OLX increase, improve the life of dealers. We have seen several examples of iFood. We are seeing more and more from Just Eat and Just Eat Takeaway going forward. That's where we are at this moment. Now you have seen a foundation, Large Commerce Models. You have seen how we use agents for our own work. We see agents for partners.
There's one more step that we need to do. How do we bring this forward? How do we make the life of all our consumers potentially easier? That's where life assistants have a role.
Did you order the shopping? You are welcome. Wow. Thank you. You're welcome.
I love all that. I think you all do, right? That's a very easy type of life. You know? It's just all these chores, they're taken care by something else and other agents and so on. It's not the way it is now, right? The reality is actually different. Everything is digital, and nothing feels simple, so we need to fix that. 15, 20, 50 apps that we use continuously every day. Hundreds of decisions, lots of noise across all this digital ecosystem. How do we change that? Well, this way. That's where we go. Something super simple. Just ask. Consider it done. I want sushi at 1:00 P.M. Got it. I will order it. Please add a Coke. Fantastic. Is that realistic? Is it far away? Not really.
Everywhere across the group, we add layers of technology in our platforms, which make the interaction with our own platform easier, smoother, based on language, seamless, more personalized that understand intent better. Ailo is one of those. It's again developed by iFood, and it has a number of features. Some of those that we have just seen takes about one minute to go from the first message to an order. It gets fast because it tells you what you probably like already, so you don't have to scroll it in there. It narrows the funnel because already thinks what you like. Sometimes it's so good, you just have one click. It tells you what you want right away, just do it, and so on. By the way, you can do it as you like. You can call, you can use voice, you can use WhatsApp.
Any other thing is fine. Let's look into specific how this works. Nishi.
Thanks, Euro. Hi, everyone. Let me introduce you to Elena. Elena lives in Buenos Aires with a partner and two kids.
She works very hard, she travels when she can, and like most of us, she juggles a lot of different priorities in life. Today, I'm incredibly excited to share with you how, with some of the life assistants we built at Prosus, will make life simple for Alina and millions of users like her. Alina loves surfing, and she has been dreaming about going to Rio for some time. Let's see if SOFIA can help her. SOFIA is our life assistant for travel with Despegar in Latin America. Alina pulls up SOFIA on Despegar app, and she expresses her intent to go to Rio and asks for suggestions. Let's do that live. Hi, SOFIA. I'm thinking long weekend in Rio for surfing. What do you suggest?
Now what's happening behind the scenes, it's pulling information about Rio, and it's making it personalized for Alina. As you see here on the screen, this is not general Wikipedia information on Rio. Alina is interested in surfing, and SOFIA is suggesting, so the life assistant is suggesting Alina surfing beaches that she can go to. Not only that, SOFIA also knows that Alina has kids, and she might be traveling with her kids. It's also suggesting surfing school for the kids that Alina can go to. It's also presenting this information in a visual way, so she can interact with it, get inspired, and explore further. Maybe at this point, Alina wants to see what other places she can visit in Rio while she's there.
Again, SOFIA is pulling this information, and again, what you see here on the screen, it's very contextual. It's not suggesting all the tourist spots in Rio, it's suggesting things considering that if she goes there with her kids. For example, the top recommendation is to go to Sugarloaf Mountain, and then some other points which she can again visually interact with and make a decision. Now, Alina doesn't take the decision on spot, right? Because that's not how we decide for travel. She thinks about it. She discusses with her partner. Maybe a few hours later or a few days later, she comes back to it and she decides to book. Let's see what happens then. This is that moment. Okay, let's do it. Help me book flights and hotel for family, let's say for 20 March, five nights.
Now, again, what's happening behind the scenes. The agent is pulling all the information, all the combinations of flight, hotel and room that is suitable for Alina. Now, what you see here on the screen. SOFIA responds with three options. If I click on it, you see three options. Think about it. These are three options, not 300 options. We have all been there. Even when you decide where you want to go and what dates you want to go to, there are thousands of combinations of hotel and flight that you need to scroll through for many hours, apply filters, you know, sort by price. Alina didn't have to do any of that. It's because SOFIA is very confident what Alina will like.
Again, this is powered by Large Commerce Model because SOFIA has a deep user understanding of Alina because of the Large Commerce Model that we discussed already. SOFIA is recommending these three places to Alina, and it's also suggesting that the second package is best for the family. Maybe in this case, it's a combination of flight and hotel, and Alina goes with it, and she can go further ahead. She can do the checkout on the Despegar app, and she can book it. Let's pause here for the moment. Let's not make this booking for Alina. She can do it later. What you saw here is from, "I want a getaway," to a confirmed booking in just few messages. No scrolling web pages, no comparison tabs. This is possible because...
This is possible when you combine agents that can execute with Large Commerce Model that knows business intent. Agents take care of execution, LCM takes care of personalization, and together they make commerce feel effortless. This experience is live already in Latin America with millions of users of Despegar. Some of the features that we showed related to personalization, they are in early preview and will be rolled out soon. What you shared with Ailo and what you see here with SOFIA, these are life assistants which is embedded in our businesses, iFood and Despegar. For us, this is the starting point for life assistant. Why stop here? Let's shift gear a bit. Let's take Alina's story further. Alina can use SOFIA to book travel. She can use Ailo to help her with food.
What about other things in life which is not travel, which is not food? Everyday matters that still take a lot of time to organize. Let's take a scenario. Maybe it's Alina. Alina wants to book a party for her kid. Can a life assistant help her? This is where Zapia comes in. Zapia is a Prosus portfolio company in Latin America. With Zapia, we have a true hyperlocal life assistant that can help Alina, and I wanna show you with a video. Let's get the video.
Okay, let's suppose I'm the user, Elena. Hey, Zapia. Quick question. Next month is my son Luke's birthday, and I'd love to organize a small celebration for him. I'm not sure where to start. Okay, Zapia immediately understands the context. She already know Luke is turning nine, that he loves Star Wars, and that there are about 30 kids in his class. Instead of asking questions, she jumps straight to a plan. Venue, food, decorations, cake. This is the first thing to notice. Zapia remembers context from previous conversations and turns a simple intention into a concrete plan. Great. Could you find kids birthday venues in Zona Norte, compare the best options, and contact them on WhatsApp to check availability for April eleventh? All right, Zapia searched for venues, identified the best options, and started contacting them on WhatsApp.
If I open WhatsApp, you can actually see the conversations happening here. Zapia is already coordinating everything. I didn't browse websites, compare reviews, or message businesses one by one. Zapia handled the discovery, the comparison, and the outreach. Perfect. If one of them is available, ask if they can handle catering and decorations too. Nice. One of the venues is available, and they can take care of the venue, food, and Star Wars decorations. Most of the party is already organized. Thanks, Zapia. Could you also find a bakery that makes a custom Baby Yoda cake and have it delivered to the venue? Zapia finds a bakery, confirms the cake, and coordinates the delivery. One more thing, since we're planning to give Luke a new bike, could you monitor the prices over the next few weeks and send us weekly updates?
Just like that, Zapia is now tracking bike prices and sending reports automatically. In just a few messages, Zapia helps organize the entire birthday, venue, cake, and even the gift. Dream to transaction. Today, it's a birthday party. Tomorrow, it can be dinner, shopping, or planning a trip. This is what we mean by an operating system for daily life. An AI layer on top connected to real services underneath. Food delivery, events, travel, payments, all connected and all aware of who you are and who you're with.
Wasn't that impressive? Zapia is live today already in Latin America. It has six million users and growing. What you just saw is Zapia Max. It's available to thousands of users, and we are rapidly rolling it out to more users. You can already download Zapia app and join the wait list for Zapia Max. If like me, you are very inspired by this video and you can't wait, you can send me a message, and I'll give you a special access code that you can use. Now as a next step, we are integrating Zapia to Prosus ecosystem services. What that means, in the same assistant, you can ask the same assistant to order food from iFood or to book travel with Despegar or to browse events in Sympla, all in same conversation.
We are very excited about these next steps, and as my colleague said, we are just getting started. Now I would like to invite Yaro back stage.
Fantastic, Nishi. I love this application. We all use it. By the way, flood Nishi with email requests. Don't worry about that. Just send him the request. I need the code. He's happy to serve that. The other thing that it should be obvious, but I think is good to underline, it's perhaps possible to create this type of assistance without having access to food delivery or to cars or other things. It's not impossible, but only when you have a deep integration with the services that the value comes out. That's the key for what we do. It's not only creating a software layer on top of it that can do different things, which are great or browse the web. It's when you have to deliver the services, you need to integrate deeply, and that's where value is.
In any case, at this moment, we have about 20 live assistants across Europe, India, and Latin America. They are Ailo, they are Sofia of Despegar. There are many others that operate for you in travel, food, personal assistant, study and learn, but also in Europe with food or OLX autos many cars for jobs, and in India, Swiggy, concierge and so on. All these will continue developing. Everything that you have seen here is really deployed, but it continuously develops and is going to become more sophisticated over time. That's, to me, it's. Oh, sorry, because I'm also forgetting here. 50 million users are already using this tool. So even if it's early on, it's already 50 million users. It's already a large community of users that use this assistant. Many highly impactful use cases are disclosed to this assistant.
Does it have revenues? Yes. Maybe not the huge amount that you expect, but they're already starting, and we just started 3%, and it's growing. These tools are going to create the next wave of growth for what we do. At this point, I'm going to stop here. I'm going to sum up what we have seen today. There are many things that we have seen, and I want to summarize them all together. First of all, we talked about Large Commerce Models, and at this moment, LCM is available for all companies in the Prosus Group for use in powering their organization, their applications. We're also working LCM for ecosystems, so we can work between companies and leverage the entire intelligence across companies in Latin America, in India, and in Europe.
This is going to be available in mid-2026, and we start with Latin America and then India. The second set of products is Toqan for Work. It's available now everywhere for all the Prosus companies, and there are hundreds of high impact agents for work which our colleagues share with each other in the marketplace. Somebody in eMAG is doing something super interesting for finance, then our colleagues in OLX can use that agent for finance. And that's the mechanism that we facilitate as much as you can. All this is available now to everybody at Prosus. Agents for partners. Dealers, we have seen that for OLX. Restaurant, we have seen several for iFood and Just Eat available now. What we are releasing in the next few months, Toqan for Partners, a platform where partners, any partner, can build their integration.
It's going to be not only for restaurants, hotel, dealers, and all our partners, but also for external companies that want to build tech on top of that for our partners. It's going to be early access in March, so soon, and it's going to be deployed first in Europe and then in Latin America. Finally, for the life assistants, you have seen several for travel, food, jobs, cars. They are all available now. They really power several of our applications and consumers. You also have seen Zapia Max. Early access. Zapia is the last one that you have seen. It's powered by Ecosystem. Early access soon. In March 2026, we start in Latin America, and remember to send the email to Nishi if you want to have early access. Before I close, there are two more things that I want to share.
The first one, there are other tools and products that we are developing, we are really excited about, but they're really early stage. This is Flume. Flume, it's a personal assistant in a proper way. Flume runs with you, runs in the background all the time. You can connect it to your life, your WhatsApp, your calendar, your email, to any other system that you want Flume to use, and it can take care of everything that you do. It's really an incredible tool. We use it internally at Prosus. We use it every day. It can help you do things. It can change the calendar. It can make appointments for you. It really is, let's say, the assistant that you imagine. We use the same technology of OpenAI, but what we have done, we have made OpenAI for the cloud.
It is secure for use at scale. That's important for us. You're probably all familiar with Open Claude, the personal agent. This is a version of that. It's in use at Prosus. Selected availability in Europe in April 2026. We'll publish when we start having whitelisted early users, and we expect to release this sometime during the year. This is an experimental program. Extremely excited about this, but it's early. Whenever we open it up, you're welcome to test. Finally, many of the things that we do, we are open sourcing them. We use a lot of technology. We develop a lot of technology in-house. We find it extremely useful. More and more is going to be open source. We start with AgentWeb and Evaluation Tool. A tool is an agent, in fact, that helps you evaluate applications that you create, maybe consumer applications or other applications.
We are releasing LongMemEval, a tool to test memory for agents. Agents work well when they remember things. You need really to find out what's the best way to remember, so we have a tool for doing that. We also spend a lot of time together with the companies within the group to develop this open protocol for end-to-end agentic transactions. Agents that buy from agents, that's the way things work. Agents need to know that we are buying complicated things, thousands of customizations in food and all these kind of things. We have to create a protocol for making that work. That's not ours. It's an open protocol. Agent service protocol. Finally, we develop a version of Open WebUI, which we call WebUI Hive, which is for multi-tenant. It means for groups. It's designed for teams to be used.
It has higher security and governance. You can use all these tools anytime. These are the QR codes. We'll publish all that material as much as we can. This is an opportunity for all of us to collaborate on things that matter for us, but hopefully also the things that matter for you, and we count on your help to make this a big thing. With that, I think it was an interesting series of releases. I think it was an interesting set of products, and I hope you enjoyed that, both you here and both you that are online. I need to give back now the stage to Fabrício to close all this.
Hello, everyone. Hope you had fun with our hour and a half of presentations. I want to tell you a funny story. When we said we are going to do that, I suggest let's do it four or eight hours. Then we can do tech and presentation of everything in details. In the end, it was a new idea. Let's be brief. Let's just give an overview and open all the data and material to you later. Hope you get inspired on what we are doing. We have a lot of more things to keep showing and sharing, and that's the way Prosus going to keep working, developing here, connected to our billion customers, but sharing the knowledge, the research with the community and with our partners. For the first time, our objective is to work with the partners to build things together.
I don't know if you got what Euro Beinat said. I love that. Through the Toqan for Partners, the LCM that you saw is going to be available not only for our restaurants and dealers, but also to other companies build on top of them. I think this is amazing because you have many companies that want to distribute to restaurants. We have millions of restaurants, and we are supplying a platform to distribute products to them and also to personalize better because we are sharing the model with these partners. I think this is great, and I think we are going to have lots of fun talking more about that over the next few month. To finish, today we talked a little about agentic workforce, and I think Prosus is quite advanced on that.
This is just getting started, as we heard a lot today. The companies are going to run over agents. We have a lot of them. We have to move fast to make sure our partners and our company are more advanced or more sophisticated than the competitors, and I think we are doing that. We talk about commerce models. Some other companies are trying to do that. I'm quite proud that what we are doing here is sophisticated and a competitive advantage, and opening that for partners, I think it is an amazing step for Prosus. We talk about life assistants. I love the Zapier demo, and as a curiosity, the Flume one. We are discussing if we're going to open for you all today or not. I really use it every day, and I say, "Flume, I'm going to travel next month.
Summarize my agenda, give me tips, and send an email to my wife about where I'm going to be." They do all of that, and sometimes they do something wrong, and I say, "No, Flume, don't do that." It's very funny. I now have two assistants, Olivia, that is seated here, and Flume, and I talk between them every day in Portuguese and English, and it keeps helping me. We decided to keep it running just inside Prosus for now. My dream, we are going to insert that in all our services. That's the direction we are going. You saw something, for example, SOFIA and Ailo that already has, like a few million users, and you saw something that is very new, and we are going to push it for a few million users as soon as this is ready for prime time.
As I told you, this is just the start. Our clear dream is agentic companies, meaning it's not only the Toqan for Work with tasks that they do every day, but it is really an owner that manage the whole business. There is a few companies starting to do that around the world. Our vision is we can do that for restaurants and dealers better than anyone else in the world, and hopefully we will keep evolving the Toqan for Partners platforms to get there. This is just the start. The future, we will have embodied agents. I saw a few embodied agents outside. Our robot walking. I thought he was around here. It's not. But you saw our robot, Darwin, in the reception. You saw our robot doing food delivery. That's the small one.
We have a much bigger one that's doing food delivery in the streets. I hope you get the authorization to launch it in Amsterdam very soon. It's already running in U.K. I want to finish with this slide to show you. One year ago, we thought maybe AGI and the disruption of AI is five to 10 years ahead. We believe the disruption and the complete transformation on how we work is in the next one to two years. We are sharing with you a few things today. We have a long list of things we are working, and I hope we can keep working with Prosus team, but also our partners to lead on the transformation that we have ahead with AI. Lots of more things ahead, but I'll save something for our next Prosus Forward in a few months.
With that, on time, what's crazy for a Brazilian. I'm feeling, like, strange maybe because I just talk 10 minutes in the beginning and five minutes in the end. If I talk the full one hour, it would take three hours, as probably you already know. On time, I'd like to say thank you for everyone. We have lots of people from Prosus team involved here in Amsterdam, but in Brazil, India, and Europe, lots of places. All of our companies participate in all those things. They are really running. We have iFood, Despegar, OLX, and more, 10 or 15 companies. We have many people watching us and partners, investors, journalists here. Thank you for coming. You are going to hear much more from us about disruption, technology, and innovation, and hope you can build an amazing future together. Thank you very much.
See you next time.