Menu

Categories

Tags

Core-Mate’s GUI Agent Raises Millions, Open-Sources Its Screen-Reading Tech

June 12, 2026 | Source: qq | AI, ByteDance | 131 views 0 comments

Core-Mate, a GUI agent execution platform built by a team of former ByteDance engineers and product managers, just closed a multimillion-yuan funding round. The investment comes exclusively from Dunhong Capital, led by partner Yu Wenchao. The fresh capital will go toward upgrading the underlying technology, iterating on the mobile-first UI visual perception capabilities, and expanding the user base.

On the technical side, Core-Mate uses a multi-source fusion approach to understand interfaces. It combines visual data, DOM node trees, control hierarchies, region segmentation, and page state changes to build more stable screen semantic recognition. The underlying screen perception framework, OpenGUI, has been open-sourced for developers. It provides task execution, process recording, and automated QA infrastructure. An Action Recorder lets users record their own workflows to cut down on token consumption, and the agent can reflect on its execution results and correct itself — basically, it’s designed to keep getting better on its own.

The founding team is largely from ByteDance. CEO Sang Yifan was a product manager for Douyin (China’s TikTok) and Doubao (ByteDance’s AI assistant). Cofounder Xu Zhenzhen led operations and growth for domestic and international businesses. Cofounder Meng Yao is a serial entrepreneur with cross-industry experience in China and overseas. The team’s take? Large language models today are like a brain without hands or feet. The key to commercial AI, they argue, is getting it into real business environments where it can reliably execute long, complex tasks for users.

Leave a Reply

Your email address will not be published. Required fields are marked *