Alibaba's Qoder Evolves into AI Task Workstation: Hands-On Review

Alibaba's Qoder has released a new desktop application that shifts from being an AI code editor to an AI task workstation. The new version removes the built-in code editor and instead acts as a resident on-device agent that plans steps, modifies files, and runs commands autonomously.

Installation & Getting Started

Download the installer from the official Qoder website; Windows, macOS, and Linux packages are available.

After launching, two work modes appear: Programming mode for development projects and General mode for daily tasks.

An Auto button under the input field lets you choose a model tier: Auto, Extreme, Performance, or Economy. Auto selects a suitable model automatically and is sufficient for daily use.

You can also manually pick a specific model such as Qwen3.8-Max, GLM-5.3, or Kimi-K3 from the model dropdown.

Qoder supports Bring Your Own Key (BYOK): go to Settings → Model → Add Model, enter your API key. Pre-configured providers include Alibaba Cloud Bailian, DeepSeek, Zhipu, and Kimi, plus custom OpenAI/Anthropic-compatible base URLs. Using your own API key does not consume Credits.

Programming Mode Development

The following demonstrates developing a Vue3 + Element Plus backend management system to test the new Qoder's development experience.

Create a workspace: click the plus icon next to Workspace, select a folder, and name the workspace; all generated code will be placed there.

Choose an approval mode: Ask, Auto, or Full Access. Auto mode is convenient for development — it only prompts for confirmation on risky operations.

Then enter the following prompt:

Build a backend management system frontend skeleton with Vite + Vue3 + Element Plus:
- Only two pages: login page and user management page. Login page has a username/password form; clicking login redirects to user management page.
- User management page includes a user list table with search box, add button, several columns of sample data, and pagination.
- After setup, help me install dependencies and start the dev server.

Qoder completed the implementation, launched the browser to run tests, and verified the result without any manual intervention. Unlike previous agents that required the user to capture screenshots and copy console errors, Qoder connects directly to the browser, observes the rendered page, and self-corrects.

The first version lacked a sidebar, which would hinder future expansion. Using Qoder's browser annotation feature, I circled the area on the page and asked it to add a sidebar. Qoder modified the code and refreshed the page automatically. Other agents would have required manual screenshots and explicit instructions.

Similarly, I used annotation to request a search button on the input field; Qoder updated the code and refreshed for verification.

The final result is a working backend skeleton with login, user management, and a sidebar.

Other Common Features

Extensions

Qoder's Extensions panel includes Plugins (e.g., Superpowers, Chrome DevTools) for one-click installation, Skills to codify repetitive tasks, and MCPs (Model Context Protocols) to connect external tools and data sources.

Appearance

The UI offers two color schemes and four themes, switchable instantly in Settings → Appearance for comfortable long-term viewing.

Data Import

If you previously used the old Qoder IDE or QoderWork, Settings → Data Import can migrate task records, memory, and MCP configurations in one click.

Computer Control

Computer Control lets Qoder view and operate desktop applications. Enable it via Settings → Computer Control → Enable Computer Control. Combined with browser connection, it can attach to an already-open Chrome or Edge instance, preserving login state and cookies, and inspect network requests and console output.

Mobile App

Qoder also provides a mobile app to monitor agent progress, handle critical approvals, add instructions, and preview results directly on the phone.

The overall experience shows that the new Qoder feels less like an AI code editor and more like an AI task workstation that can complete tasks end-to-end. From generating project scaffolding, invoking browser tests, to annotating UI changes for automatic code updates and verification, the workflow required minimal manual screenshot capturing. Flexible model tiers, BYOK support, computer control, and mobile access make it a comprehensive tool for daily development.