Kakao Unveils Kanana Space AI Assistant for KakaoTalk
Kakao introduced Kanana Space, an integrated AI assistant designed to understand spoken requests and perform personalized tasks inside KakaoTalk, with a closed beta test planned for December.
Kakao has unveiled Kanana Space, an integrated artificial intelligence assistant that the South Korean technology company plans to bring to its KakaoTalk messaging platform after a closed beta test in December.
The company introduced the service on Tuesday at its annual developer conference at Kakao’s AI campus in Yongin, Gyeonggi Province. Kakao said the assistant is designed to understand verbal requests and act on the user’s intended purpose rather than requiring a series of conventional on-screen commands.
Chief Executive Chung Shin-a described the project as a shift from a graphic user interface toward what Kakao calls a language user interface. Under that approach, users would communicate naturally while the service interprets context and carries out tasks. The strategy places conversational interaction at the center of Kakao’s effort to integrate AI more deeply into its flagship messenger.
A demonstration showed Kanana Space using several layers of AI to provide personalized assistance. The service could recommend workout clothes after considering both weather conditions and a user’s exercise data. It could also organize schedules by interpreting information contained in a user’s KakaoTalk message history.
Kakao also outlined the technology behind Kanana, its foundation AI model that powers the new assistant. The company said its system is designed to select an appropriately sized model for each request rather than relying on its largest model every time. That approach is intended to deliver strong results while using computing resources more efficiently.
To preserve the performance of smaller language models, Kakao said it applied Multi-Teacher On-Policy Distillation, a post-training method that transfers capabilities from multiple teacher models. The architecture reflects the company’s effort to balance response quality, speed and operational cost as it prepares the service for wider use.
Chung said the standard for evaluating generative AI is moving beyond whether a system can produce good answers and toward whether it can complete requested tasks. Kakao’s goal, she said, is to add intelligence that recognizes individual differences to a platform already built around connecting people.
The December closed beta will provide an early test of whether Kanana Space can translate that ambition into reliable everyday assistance. Its eventual integration into KakaoTalk would place the service inside one of South Korea’s most widely used digital platforms, giving Kakao a direct route to introduce task-oriented AI to a large consumer audience.