Skip to main content

Marvis Now Lets You Plug In Your Favorite AI Models

On September 1st, Tencent's operating system-level AI assistant, Marvis, rolled out a new "Custom Model" feature that's turning heads. The update lets users plug in third-party models that follow industry-standard interface protocols—think Tencent Cloud, Alibaba Cloud, DeepSeek, Minimax, Kimi, Zhipu, Xiaomi, and more. You can even connect open-source models running on your own machine, and your settings sync across all your devices.

The Freedom to Choose Your AI Brain

Over the past year, the large model race has been moving at breakneck speed—new champions seem to emerge every month. No single model excels at everything. Some shine in long-form writing, others in code generation, and still others in speed and cost. The desire to switch between models on the fly has grown louder.

This update directly answers that call. Before, Marvis users were stuck with the platform's built-in models. Now, you can pick a pre-configured combo like HuanYuan Hy4preview and DeepSeek-V4Pro, or connect any other model that speaks the industry-standard language.

Getting started is straightforward: head to Settings > Model Management. The system already lists quick configs for models from Tencent Cloud, DeepSeek, and others, but you can also add your own custom entries. Once set up, you can switch models mid-conversation, picking the best tool for the task at hand.

Image

A Generous Daily Quota—But There's a Catch

Marvis gives every user a free daily quota of 10 million tokens—plenty for everyday use. But if you're a power user cranking out code or processing long documents, that quota might run dry. Here's the workaround: when you access third-party services through custom models, the usage is billed to your own account, not Marvis's quota. So heavy users can bypass the free limit entirely.

Local Models: Privacy Without the Hardware Headache

For those who value data privacy or want to cut costs, this update also opens the door to local models. You can run open-source models on your own computer via third-party apps, then configure them in Marvis to use in "efficiency mode." That's a game-changer because Marvis's previous "local mode" demanded beefy hardware. Now, you can enjoy local processing without needing a top-tier GPU.

Plus, your conversations stay on your device—no cloud servers involved. That's a big win for anyone handling sensitive data.

One Setup, Every Device

Another standout feature is cross-device sync. Your custom model configurations are tied to your account, not a single device. So whether you're on Windows, Mac, iOS, Android, or iPad, your model list follows you. Set it up on your office computer today, and it's ready on your phone during tomorrow's commute. No more re-adding models on each device.

The Bigger Picture

Decoupling AI assistants from their underlying models is becoming the industry norm. Users don't want to switch apps just to try a new model, nor juggle multiple assistants. Marvis's custom model feature is a step toward that flexible future.

The feature is live in the latest Marvis update. If you haven't tried it yet, head to Settings > Model Management and see what fits your workflow.

Key Points:

  • Marvis now supports custom models from major providers like Kimi, Zhipu GLM, and DeepSeek.
  • Users can also integrate locally deployed open-source models for privacy and cost savings.
  • A daily free quota of 10 million tokens is provided, with third-party usage billed separately.
  • Custom model configurations sync across all devices via your Marvis account.
  • The feature is available in the latest version under Settings > Model Management.

Image