Alibaba's Qoder Coding Agent Gets a Smart Glasses Client

Alibaba announced a glasses edition of its Qoder coding platform on September 3, 2026, running on Qwen and Rokid AI glasses with full-duplex voice.

by HowAIWorks Team
On this page

Introduction

On September 3, 2026, Alibaba Cloud announced Qoder Glasses Edition, a wearable client for Qoder, the agentic coding platform it launched in August 2025. The first supported devices are Alibaba's own Qwen AI Glasses and Rokid AI Glasses — sold in China under the Leqi brand. The Qwen version has entered directed beta testing, with the complete feature set promised for the 2026 Apsara Conference in Hangzhou on September 22–24.

The pitch is not that you write code through eyewear. It is that the glasses become a hands-free control surface for an AI agent that is already doing long-running work somewhere else.

How the Work Is Split

Alibaba describes a three-client division of labor, and it is the part of the announcement worth reading closely:

  • Desktop keeps the project context and executes the complex work — the repository-aware agent stays where the repository is.
  • Mobile is for checking on tasks remotely.
  • Glasses contribute what the other two cannot: a voice channel and a first-person view of the physical world.

So the compute and the codebase never leave the machine. What the glasses add is presence — the ability to stay in the loop with a running task while away from the keyboard.

Full-Duplex Voice Instead of Wake Words

The interaction runs over a full-duplex voice channel, meaning both sides can speak at once rather than trading turns after a wake word. Alibaba's framing is talking to the agent the way you would talk to a colleague: hand over a task, ask how the build is going, add a requirement mid-run without restarting the exchange. What implements it is not disclosed — end-to-end speech-to-speech models are the usual route to that behavior today, but Alibaba says only "full-duplex voice solution."

For approvals, the model is visual rather than conversational. When the agent hits a high-risk operation or a checkpoint, the relevant information appears as a card at the edge of the field of view, and you confirm by voice or by tapping the temple of the frame. Task completions, exception alerts and progress updates arrive the same way, and Alibaba says none of it interrupts whatever the agent is currently running.

The Camera as a Debugging Input

The most concrete claim is about computer vision rather than voice. On command, the glasses capture what you are looking at, and Qoder analyzes it for code errors, device status, whiteboard sketches and screen content — then correlates that against the project it is working on and proposes next steps.

That targets a real gap. An agent running in a terminal — Claude Code, Cursor in agent mode, or Qoder's own CLI — cannot see the error on a second monitor, the status light on a rack, or the architecture someone just drew on a whiteboard. A multimodal capture is a cheap way to get that context into the loop without retyping it.

What Is Not Yet Known

Alibaba has published no latency figures, no pricing for the glasses tier, no accuracy numbers for the visual recognition, and nothing about availability outside China — the two supported devices are both Chinese-market products. There is also no primary source in English, or on the web at all: at the time of writing neither Qoder's blog nor its changelog mentions the release. The announcement went out through Alibaba Cloud's Chinese channels, and every account below — including this one — traces back to that single post. Treat the feature descriptions as vendor claims relayed at one remove.

Conclusion

Qoder Glasses Edition is a distribution move more than a capability one. The agent is unchanged; what changes is where you can be while it works, and how much of the physical world it can see. Whether developers want a coding agent talking into their ear is an open question — but the visual-context piece addresses a genuine limitation of terminal-bound coding agents, and it is the part worth watching when the full version ships at Apsara.

Sources

Frequently Asked Questions

It is a wearable front end for Qoder, Alibaba's agentic coding platform. The glasses supply voice and a first-person camera feed; the desktop client still holds the repository context and runs the actual work.
The first two supported devices are Alibaba's own Qwen AI Glasses and Rokid's glasses, sold in China under the Leqi brand. The Qwen version is the one currently in invite-only beta.
Not by dictating source. The interaction model is assigning tasks, checking build status, answering the agent's questions and approving steps by voice while Qoder's desktop client does the editing and execution.
Capturing things that are not on your screen. Alibaba says the assistant can read error codes and screen content, device status indicators and whiteboard sketches from a first-person capture, then correlate them with the project it is working on.
Directed beta testing opened with the September 3, 2026 announcement. Alibaba says the full feature set ships at the 2026 Apsara Conference, held in Hangzhou from September 22 to 24.

Continue Your AI Journey

Explore our glossary and model catalog to deepen your understanding.