[Edaily Reporter Lee So-Hyun ] Kakao(035720)KakaoTalk is transforming from a messenger that connects people into an “integrated agent-based AI interface” that understands user requests and performs actual tasks. Starting with “Kanana Space”—which allows users to summon the service from anywhere in KakaoTalk to send messages and manage schedules—the platform will expand to a system where users can handle orders, gifts, reservations, and money transfers with a single voice command.

Kakao unveiled this AI strategy on the 13th at “if(kakao)26,” held at the Kakao AI Campus in Yongin, Gyeonggi Province. The slogan for this year’s event is “Intelligence in Every Connection.”

Jeong Shin-ah, CEO of Kakao, who served as the first keynote speaker, stated, “We aim to take the ‘connections’ Kakao has built over the past 15 years to a new level,” and announced plans to integrate AI—which understands individual preferences and circumstances to perform necessary tasks on their behalf—across the entire KakaoTalk platform.

Swipe the KakaoTalk Screen and ‘Ryan’ Appears… AI Handles Messages and Schedules

The first service unveiled was “Kanana Space,” an AI assistant based on KakaoTalk.

When users need help while using KakaoTalk, swiping the screen with two fingers brings up the character “Ryan.” Users can ask Ryan to send messages via voice or text, manage schedules, perform web searches, and change KakaoTalk settings. It’s also possible to have casual conversations with Ryan.

Kakao plans to expand the scope of Kanana Space to include orders, gifts, reservations, and money transfers in the future.

For example, if a user says, “Order some bottled water for me” during a conversation, the service will reference past order history to select the product and even apply available coupons, handling all necessary steps from ordering to just before payment.

Kanana Space will begin a closed beta test (CBT) this December for a select group of pre-registered users. Applications for participation will be accepted on a separate website starting on the 13th.

Kim Dae-nyeon, Head of Design at Kakao, said, “We aim to provide an experience where anyone can easily benefit from AI assistance without having to learn new ways to use it.”

Create ‘Your Own AI Employee’ on Your PC… Kanvas to Launch CBT by Year-End

The company also unveiled “Kanvas,” a service that allows users to create their own AI agents directly on a PC.

When a user describes the desired agent in natural language, Kanvas connects the appropriate tools and skills to configure the functions needed for execution. The created agent performs tasks within the scope authorized by the user and seeks the user’s confirmation only at critical stages.

Users can delegate repetitive tasks or daily chores to the agent, and the company plans to introduce a feature in the future that allows users to share the agents they’ve created with others.

Kakao will conduct a closed beta test (CBT) of Kanvas for a small group of users as early as this year.

Ultimately, the goal is to create a system where users can summon and utilize different AI agents all within a single KakaoTalk app.

In the first half of next year, Kakao plans to introduce a new space within KakaoTalk for connecting personalized agent teams, and it intends to support the integration of agents created not only with Kanvas but also with those built on other frameworks.

From On-Device to Large-Scale AI: ‘Kanana Full-Stack’

Behind the service lies Kakao’s proprietary AI model, “Kanana.” Kakao has identified the following as its core technological directions: △ full-stack optimization encompassing lightweight design, computation, and serving; △ multimodal capabilities that understand text, images, and speech together; and △ agentic AI that translates user requests into actual actions.

The architecture is designed so that simple requests are handled by on-device models running on smartphones, while complex tasks are connected to large-scale models.

“AI Conductor” analyzes user requests to determine which models and tools to utilize and in what order, thereby optimizing costs and processing efficiency.

The company also unveiled “Kanana-o,” an omnichannel language model that understands text, images, and speech simultaneously and responds via voice. Kakao developed the encoder and decoder that make up this model in-house and plans to release them as open source in the future.

The company also showcased its development of “Kanana Live,” a full-duplex voice technology that allows users to speak while simultaneously listening to the other person.

“Project Mosaic”: Connecting User Memories

The company is advancing “Project Mosaic” as the foundation for personalized AI.

This structure connects users’ scattered memories and experiences into a single context, while ensuring that each agent receives only the information necessary to perform its tasks, within the scope permitted by the user.

The goal is to enable multiple agents to understand the same user context and collaborate effectively.

Kakao is also establishing a data governance and safety verification system. This system verifies that the action plans established by agents and their actual execution remain within the scope agreed upon by the user.

CEO Jeong stated, “We will build upon ‘Connection for All’ by adding ‘intelligence’ that understands each individual and works according to their specific needs,” adding, “We will ensure that everyone can experience AI in their familiar daily lives.”

Meanwhile, at this year’s if(kakao)26, employees from Kakao and its affiliates will lead a total of 55 sessions on topics including AI direction, trust and safety, models and Agent AI, and infrastructure.