Skip to main content
We believe the Agent is evolving into the next-generation operating system (OS). In this paradigm shift, a plugin is no longer just a feature extension: a plugin is software, and a plugin is an app (a Mini App). At the same time, real productivity in every industry remains rooted in highly specialized work environments. A Chat UI alone can’t replace the precise control and complex views of professional tools. Future professional workflows shouldn’t ask users to leave their workbench to accommodate AI; professional work environments and AI should grow together, with AI built in as a foundational capability that can be called on at any time, like utilities, compute, and storage. By building professional-grade Mini Apps that grow on top of the Agent OS, we aim to connect the specialized interfaces of each vertical with AI-native compute underneath, and reshape how every industry works.
Mini App at a glance

You and the Agent work in the same Mini App; the host provides the runtime and permissions.

​
What can Mini Apps enable?

  • Analysts can gather stock information scattered across Zsxq (知识星球), Jinmen Finance (进门财经), and Tencent Docs (腾讯文档) into one dashboard. The Agent sees everything on it and can carry the work forward at any point.
  • Operators can turn tasks in Notion or Jira into a list they create, edit, and delete on the page, with changes synced back.
  • A team of bots can split up work in a Mini App: a lead bot hands long tasks to background Agent sessions, then reviews and reports on the results. MiniMax Bot works this way.
  • A road trip planner chains Amap (高德地图) routing, Xiaohongshu (小红书) search, and image generation into one itinerary workspace.
  • You can describe what you need in chat and have the Agent build a Pomodoro timer, an expense tracker, or a monitoring page on the spot, then ask it to adjust the app as your habits settle.

​
Why do Mini Apps matter?

  • Users: work no longer scatters across a pile of conversations. Data, actions, and progress stay on a page that keeps running. When page content goes to the Agent, it lands in your message input first, where you edit and confirm it; AI never presses Send for you.
  • Developers: no runtime to build from scratch. The host handles processes, ports, authentication, credentials, and publishing that rolls back, so you write only the page and business logic, or ask the Agent to write them.
  • The Agent: gains a workspace that keeps state and can be called directly. A Mini App offers its business capabilities as MCP tools, and it can hand tasks to the Agent to finish in the background.

​
How is this different from an HTML page the Agent generates?

An HTML dashboard generated by the Agent can bring scattered information together, but it only shows the data from the moment it was generated. To keep it current, push tasks forward, or save results, you have to go back to chat and explain everything again. A Mini App adds three things:
  • A backend that keeps running: the Node service starts on demand, keeps fetching data, and saves results; its data directory survives restarts, updates, and upgrades.
  • Two-way calls with the Agent: the Agent can call the Mini App’s tools, and the Mini App can ask the Agent to work in the background.
  • Permissions the host enforces: the host keeps account credentials, capabilities work only when declared in the manifest, and the service port is open only to the page and the Agent.

​
Where Mini Apps run

Mini Apps run in local sessions of the MiniMax Code desktop app (macOS / Windows); cloud, SSH, and Remote Control sessions never start Mini Apps on your machine. Install one from the plugin marketplace, download one from the open-source community, or have the Agent build one for you. Download MiniMax Code: global · China.

​
Start building

Build with the Agent

Use /miniapp-creator to go from one sentence to a working page.

Package and manifests

The plugin manifest, the Mini App declaration, and the two runtime payloads.

Write the Node backend

Export start(context); the host handles ports, lifecycle, and authentication.

Work with the Agent

Managed MCP, handing off to the message input, and starting Agent sessions.

​
Learn more

Architecture overview

The participants, the layers, and one full call from start to finish.

Mini Apps and sessions

How the service is shared, where pages open, and where the Agent can use it.

Security model

What a Mini App can access, and how the host restricts ports, credentials, and the Agent API.

Features

Interface, data, Agent collaboration, sessions, and operations on one page.

​
Community

Open-source community

Browse community Mini Apps, or share yours through a pull request.

Submit and share

Enterprises submit to the marketplace; individuals contribute to the community.