Tencent's marvis unveils custom model integration for enhanced AI flexibility

On September 1, 2026, Tencent's AI assistant Marvis introduced a 'Custom Model' feature, enabling users to integrate third-party models like Kimi and Zhipu's GLM via industry-standard protocols. The update allows users to select from pre-configured combinations or connect to other large models, enhancing flexibility in model choice. Marvis now supports both cloud and local models, with users able to run open-source models on their devices for privacy or cost savings. The platform also offers cross-device synchronization, ensuring model configurations are accessible across Mac, Windows, iOS, and Android devices. Users receive a daily token quota of 10 million, though heavy usage scenarios may require third-party integration to bypass limits. This update aligns with the industry trend of decoupling AI assistants from underlying models, offering users greater control without switching applications. The feature, available in the latest Marvis version, simplifies access through 'Settings - Model Management,' with quick setup options for major providers. By allowing local model usage in 'efficiency mode,' Marvis reduces hardware demands while maintaining data privacy. This marks a significant step in making AI workflows more adaptable and user-centric across multiple devices.