← all repositories
QwenLM/Qwen-MM-Plugins

Giving your text agent eyes, ears, and a CAD workstation

A modular plugin suite that gives existing agent harnesses native vision, video memory, audio understanding, and even CAD and 3D modeling capabilities.

2.8k stars HTML AgentsLanguage Models
Qwen-MM-Plugins
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does Qwen-MM-Plugins extends compatible AI agent harnesses with a set of multimodal toolsets. Each capability—such as reading documents, analyzing long videos, transcribing audio, or driving Blender and FreeCAD—ships as a model skill plus an optional MCP server, so the agent knows the tools exist and can invoke them. The project targets several popular harnesses including Claude Code, Codex, and Gemini CLI.

The interesting bit The core capability handles dynamic-resolution ingestion: images, video frames, and document pages are auto-scaled to the vision model’s patch grid, so a 4K screenshot and a tiny thumbnail both arrive at useful detail without manual resizing. More unusually, the Blender and FreeCAD capabilities are thin clients that command a running instance of the actual desktop software, giving the agent real parametric CAD and 3D rendering tools rather than just generating scripts.

Key highlights

  • Seven distinct capabilities, from OCR and web search to long-video memory and video editing.
  • core reads images, videos, documents, and 3D models with dynamic resolution scaling.
  • blender and freecad plug into running desktop applications via thin clients (22 and 14 tools respectively).
  • edu-agent is a skill-only module that generates step-by-step Chinese explainer videos from math or science problems.
  • Guided installer handles setup across multiple harnesses, though several cookbooks are still marked TBD.

Caveats

  • Several capabilities—including video-memory, omni-av, video-edit, blender, freecad, and edu-agent—lack finished cookbooks at the time of writing.
  • API-based features require a DashScope key; web search needs Serper. Native reading does not.
  • Windows support is limited to WSL2; native Windows is explicitly not validated.

Verdict Developers already using agent harnesses like Claude Code or Codex who need their assistant to actually look at files, watch videos, or manipulate 3D models should evaluate this. If you are not running one of the supported harnesses or do not need multimodal tool use, it is likely overkill.

Frequently asked

What is QwenLM/Qwen-MM-Plugins?
A modular plugin suite that gives existing agent harnesses native vision, video memory, audio understanding, and even CAD and 3D modeling capabilities.
Is Qwen-MM-Plugins open source?
Yes — QwenLM/Qwen-MM-Plugins is open source, released under the Apache-2.0 license.
What language is Qwen-MM-Plugins written in?
QwenLM/Qwen-MM-Plugins is primarily written in HTML.
How popular is Qwen-MM-Plugins?
QwenLM/Qwen-MM-Plugins has 2.8k stars on GitHub.
Where can I find Qwen-MM-Plugins?
QwenLM/Qwen-MM-Plugins is on GitHub at https://github.com/QwenLM/Qwen-MM-Plugins.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.