Marvis Unveils Custom Model Function, Integrating Kimi and Zhipu GLM

On September 1st, Marvis, Tencent’s operating system-level AI assistant, officially rolled out its "Custom Model" feature. This update allows users to seamlessly integrate third-party models that adhere to industry-standard interface protocols into Marvis. Supported providers include major platforms such as Tencent Cloud, Alibaba Cloud, DeepSeek, Minimax, Kimi, Zhipu, and Xiaomi. Additionally, the feature supports locally deployed open-source models and enables cross-device synchronization for a consistent experience.

Unlocking Model Flexibility: Freedom to Choose

Over the past year, the large language model landscape has accelerated, with new leaders emerging monthly. No single model dominates all use cases; some excel in long-form writing, others in code generation, and some in speed or cost-efficiency. Consequently, user demand for the ability to switch between models freely has grown significantly.

This update addresses that need by granting users full control over model selection. Previously, Marvis was limited to its built-in models. Now, users can opt for pre-configured combinations like HuanYuan Hy4preview and DeepSeek-V4Pro, or connect to any large model compatible with standard interface protocols.

From a user interface standpoint, accessing "Custom Model" settings is straightforward: navigate to "Settings - Model Management." The system offers quick setup options for models from Tencent Cloud and DeepSeek, while also providing a manual entry for users to add specific models. Once configured, users can switch between different models during conversations, selecting the most appropriate tool for each task.

Notably, Marvis provides each user with a daily free quota of 10 million tokens, which suffices for most everyday tasks. However, for high-frequency or heavy-use scenarios like extensive code generation or long-document processing, this limit may be reached. By integrating third-party services via custom models, usage is billed to the user’s external account, bypassing Marvis’s daily quota. This offers a practical solution for power users who need higher capacity.

Local Model Integration: Simplifying Private Deployment

Beyond cloud-based options, this update also supports local model integration. For users prioritizing data privacy or seeking to reduce API costs, Marvis now allows them to run open-source models locally on their computers. By using third-party applications and completing a simple configuration within Marvis, users can operate in "local mode."

The key value here is accessibility. Previously, Marvis’s native "local mode" required high-end hardware with substantial local computing power. By accessing local models through the custom model feature, users can leverage "efficiency mode," which lowers the hardware barriers and reduces performance constraints associated with running models locally.

From a security perspective, local model integration ensures that conversation data never leaves the user’s device, eliminating the need to transmit information through cloud servers. This makes it ideal for environments with strict data privacy requirements.

Cross-Device Synchronization: One Configuration, All Devices

The third major feature is cross-device synchronization. Custom model configurations are treated as personal account assets, tied to the user’s account rather than a specific device. This means that whether you log in to Marvis on a Windows PC, Mac, or mobile device, your configured model list will automatically sync, eliminating the need to reconfigure settings on each device.

Marvis currently supports Mac, Windows, iOS, Android, and iPad. For professionals who frequently switch between devices, this ensures continuity: a model configured on an office computer can be used on a mobile device during travel, maintaining an uninterrupted AI workflow across all platforms.

Decoupling AI assistants from underlying models is becoming an industry standard. Users no longer need to replace entire applications to test new models or juggle multiple assistants. The custom model feature is now available in the latest version of Marvis. Users can access it by updating the app and navigating to "Settings - Model Management."