*This is the fifth instalment of Swipe, a series in which we open up and dissect AI apps and services making waves in South Korea. This week's subject is Kanana, Kakao's AI service — which, as it happens, has just undergone a dramatic change of direction.*

Introduction

In May last year, Kakao — one of South Korea's largest internet conglomerates — launched a closed beta test of Kanana, a conversational AI agent app, staking its claim in the standalone AI services market. Designed to understand the context of both individual and group chats and offer relevant responses, the app comprised two components: Nana, a personal AI companion, and Kana, a group AI companion. This week, on 15th September, Kakao announced that the independent app would be shut down on 15th October — just sixteen months after launch.

From independence to absorption: a strategic pivot

During its brief life, Kanana accumulated a broad range of features: voice-based conversation, an AI Studio for generating images (including group photos), and niche tools such as booking car maintenance appointments. Yet Kakao has now abandoned the experiment of building a separate user base around a standalone app. Instead, it is shifting the centre of gravity of its AI strategy toward products already in commercial operation: Kanana in KakaoTalk, ChatGPT for Kakao, and PlayMCP. The company says it is closing the app in order to move to the next stage — one that draws on the technical capabilities and user feedback accumulated through operating Kanana to deliver a deeper AI experience.

Core features: AI moves inside KakaoTalk

In practical terms, the focal point of Kakao's AI strategy is now "Kanana in KakaoTalk," known colloquially as KaInTalk. It began as a beta test for a subset of iOS users in mid-October last year, expanded to Android, and officially launched on 17th March this year. It is accessible directly from an AI section within the Friends tab — the first screen users see when they open KakaoTalk.

KaInTalk's defining characteristic is its use of an on-device AI model that reads conversational context and proactively suggests relevant functions. It delivers a daily morning briefing of the user's schedule and, during KakaoTalk conversations, assists with information searches, product recommendations, and venue suggestions. In practice, asking it to recommend day-trip destinations near Seoul yields results such as Hwadamsup Forest, Dumulmeori, and Achimgoyo Arboretum, complete with nearby cafés and restaurants; asking about restaurants in a specific neighbourhood produces AI-summarised reviews and information drawn from KakaoMap. A June update added the ability to analyse and summarise chat-room conversations by time period, along with features for accommodation and travel bookings, expense splitting, navigation, and KTX train reservations.

Strengths and limitations

The on-device approach carries a clear security advantage: data is processed on the handset rather than sent to a remote server. Equally compelling is accessibility — because KaInTalk lives inside KakaoTalk, an app already used daily by tens of millions of South Koreans, users need not download anything new to gain an AI assistant.

The limitations are just as clear. One review published in April described the service as "almost there, but not quite" — a verdict borne out by documented errors, including an instance in which the AI, asked for restaurant recommendations in a particular neighbourhood, simultaneously returned results from Gangnam in Seoul, Pocheon in Gyeonggi province, and Gwangalli in Busan. This reflects a structural constraint inherent to on-device models: to run efficiently on a handset, they must be compressed, which inevitably sacrifices some capability. Parsing the relevant details from an ongoing human conversation — without the benefit of a directly typed prompt — demands a level of comprehension that is considerably higher than that required by a conventional AI chatbot.

Assessment

Kanana's trajectory offers a revealing case study in how South Korea's big tech companies are positioning AI services. Rather than attempting to attract new users to an independent app, Kakao has pivoted to embedding AI within KakaoTalk, a super-app with an already overwhelming user base. The underlying logic seems straightforward: it is more efficient to layer AI on top of a proven platform than to build an entirely new service from nothing.

★★★☆☆ (3.0/5.0)

In a sentence "It gave up on going it alone — but by returning to where it started, it gets to meet far more people."