Sakana Chat Adds Browser Code Execution
The Japan-focused Namazu assistant now turns Japanese instructions into runnable apps, games, charts and documents without requiring a login.
A Japanese assistant moves beyond question answering
Tokyo-based Sakana AI has released a major update to Sakana Chat, adding a new generation of its Japanese-focused Namazu model and the ability to execute generated code inside the browser. The service remains free and can be used without creating an account, although registered users receive conversation history and higher usage limits. Sakana Chat is currently available only from within Japan, according to its FAQ.
The new execution environment allows users to describe an application, utility or game in Japanese and then run the generated result directly in the chat interface. Sakana’s examples range from educational worksheets and simple browser games to spreadsheet analysis, chart creation and presentation materials. That combination shifts Sakana Chat from a localized conversational assistant toward a lightweight agent workspace capable of producing and testing artifacts.
Namazu is Sakana’s effort to adapt strong foundation models to Japanese language, culture and local requirements while retaining broad reasoning and coding performance. Sakana Fugu, by contrast, is a separate commercial API product that dynamically selects and coordinates multiple underlying models. Sakana’s current public product pages do not establish that Fugu itself has become the model or orchestration layer inside Sakana Chat, so the two products should not be described as a single rollout. The Chat update also did not include a new standalone Namazu weight release and is therefore a service launch rather than an open-weight release.
Why it matters
Most coding-agent products are designed first around English-language developer workflows. Sakana is instead packaging executable generation for Japanese users, including consumers, students and office workers, with no-login access reducing the initial barrier to experimentation. Its spreadsheet and document examples also place it closer to general knowledge-work agents than to a conventional chatbot.
The update is significant less because browser code execution is unique than because Sakana is combining it with Japan-specific post-training in a public product. Adoption and reliability will show whether localized agent systems can build durable advantages against global assistants that support Japanese but are not designed around Japanese workflows and cultural context. Fugu’s separate multi-model API represents another part of Sakana’s product strategy, but it should be evaluated independently from Sakana Chat’s confirmed feature set.