ByteDance Launches Internal Beta for Coze Space AI Agent Platform
ByteDance, the parent company of TikTok, has officially launched the internal beta version of Coze Space, a new AI-powered productivity platform designed to facilitate collaboration between human users and AI agents within the workspace. Built upon ByteDance’s proprietary Doubao large language model, specifically the Doubao 1.5 Pro variant, the platform supports the Model Context Protocol (MCP). This feature allows for seamless integration with existing enterprise tools such as Lark and Amap. Coze Space introduces a general-purpose agent interface alongside an 'Expert Agent' system capable of handling diverse tasks, from answering queries to executing complex work assignments. The development team initially evaluated external models, including DeepSeek-R1, but encountered limitations regarding tool invocation capabilities. After testing six different Chinese large language models, ByteDance selected its in-house Doubao suite due to its superior instruction following, multimodal capabilities, efficient tool usage, and low inference costs, which are deemed essential for scaling the platform effectively. This launch marks a significant step in ByteDance's strategy to enhance workplace productivity through advanced artificial intelligence solutions, leveraging its own technological infrastructure rather than relying on third-party models.
Wire timeline
ByteDance Launches Internal Beta for Coze Space AI Agent Platform
ByteDance, the parent company of TikTok, has officially launched the internal beta version of Coze Space, a new AI-powered productivity platform designed to facilitate collaboration between human users and AI agents within the workspace. Built upon ByteDance’s proprietary Doubao large language model, specifically the Doubao 1.5 Pro variant, the platform supports the Model Context Protocol (MCP). This feature allows for seamless integration with existing enterprise tools such as Lark and Amap. Coze Space introduces a general-purpose agent interface alongside an 'Expert Agent' system capable of handling diverse tasks, from answering queries to executing complex work assignments. The development team initially evaluated external models, including DeepSeek-R1, but encountered limitations regarding tool invocation capabilities. After testing six different Chinese large language models, ByteDance selected its in-house Doubao suite due to its superior instruction following, multimodal capabilities, efficient tool usage, and low inference costs, which are deemed essential for scaling the platform effectively. This launch marks a significant step in ByteDance's strategy to enhance workplace productivity through advanced artificial intelligence solutions, leveraging its own technological infrastructure rather than relying on third-party models.
TechNode