DuMate is the general-purpose office agent that Baidu pushed hard at its 2026 Create developer conference, and many have called it "the Chinese version of Codex." It bundles "state a request → authorize → execute → deliver" into one complete pipeline, letting users who don't write code drive AI with natural language through cross-app, scheduled, file-level office tasks — and it claims its Harness framework cuts credit consumption for similar tasks by 75%. This tutorial strings together the full getting-started path, from installation and learning the interface to two real office scenarios, suited to developers and office users who want to offload repetitive work like daily reports, idea triage, and cross-app data lookups.

DuMate supports Mac, Windows, and mobile; download it and sign in with a phone number to start using it.

Before You Begin

• Device: a Mac or Windows desktop, or mobile (this article demos on Mac)
• Account: phone-number registration is enough; the free tier works
• Download: visit dumate.cn in a browser and pick the version for your platform
• Optional connections: app accounts such as Feishu / DingTalk / Chrome / Trip.com (for cross-app tasks)
• Estimated time: 3 minutes of basic setup; about 20 minutes to run both hands-on scenarios

Step 1: Install in 3 Minutes

1. Download the Mac dmg installer from dumate.cn.
2. Double-click to open it and drag the DuMate icon on the left into Applications on the right.
3. Launch, sign in with your phone number, and complete verification.

The Mac installation screen; dragging DuMate into Applications completes the setup.

💡 Tip: Right after signing in, assign a dedicated folder as your workspace in the chat box's "Set Workspace" option, following the principle of least privilege so the agent cannot overstep into other directories.

Step 2: Get to Know the Interface — Four Core Modules

The main screen is a left-hand task bar plus a central conversation area. The four modules in the left task bar determine what you can ask the agent to do:

• New Task: the main conversation area; state requests in natural language, with voice, file upload, and multi-task parallelism (tasks keep executing in the background — no babysitting required).
• Skills: the skills plaza ships built-in skills such as Baidu Search, Baidu Miaoda, and Baidu Netdisk, and also supports custom imports via zip archive or URL.
• Connected Apps: hook up local or curated apps such as Chrome, Feishu, DingTalk, GitHub, and Trip.com so the agent can work across apps.
• Automation: scheduled tasks — for example, "summarize ten fresh AI news items every morning at eight"; it recognizes the intent and creates the task automatically.

The four left-side modules (New Task, Skills, Connected Apps, Automation) are the agent's capability entry points.

Workspace and the Principle of Least Privilege

Click "Set Workspace" in the chat box to assign the agent a dedicated folder. Inside the authorized folder, the agent has full read, edit, and create control; delete operations trigger a second confirmation. This detail makes long-term use safe — the thing you fear most with AI tools is having them touch what they shouldn't without your knowledge.

DuMate ships with a built-in security sandbox and double-confirms before deleting; the boundary discipline is well done.

Visualizing Task Execution

After you state a request, the agent shows three things at once in the right-hand panel:

• Task progress: each subtask gets checked off in real time
• Task files: intermediate artifacts and the final deliverable
• Invoked capabilities: lists the connectors used (e.g., Baidu AI image generation)

Every step is laid out in plain sight, which makes follow-along debugging smooth.

Step 3: Scenario 1 — DuMate + Feishu as an Idea-Collection Assistant

This scenario demonstrates cross-app work. Goal: have the agent visit a Feishu group on a daily schedule, collect the discussion, and organize it into structured idea cards.

1. Authorize Feishu under "Connected Apps" on the left and choose the group or document to read.

2. Back in the conversation area, enter the request:

   Collect the past week of discussion in the Feishu "Product Ideas" group, group it by topic, distill each item into a one-sentence summary, output it as a Markdown table, and save it to inspiration.md in the workspace.

3. The agent breaks this into subtasks — "pull messages → cluster topics → generate summaries → write the file" — and executes them one by one.

4. When it finishes, preview and download inspiration.md directly in the task-files area.

The execution panel for a cross-app task; you can watch the Feishu messages being pulled, clustered, and written to a file in real time.

💡 Tip: If you want this task to run every day automatically, just say "turn this into an automation task for every morning at eight" and the agent jumps to the "Automation" module to configure the schedule for you.

Step 4: Scenario 2 — DuMate + Baidu Miaoda for a Mini Game

This scenario shows how Skills play. Goal: generate a playable mini game from one sentence and get a live URL back.

1. Invoke the "Baidu Miaoda" skill from the skills plaza.

2. Enter the request:

   Build a Tetris mini game with scoring and a high-score record, and give me a link I can open directly.

3. The agent calls the Miaoda skill to generate the code and deploy it, returning a live URL a few minutes later.

4. Open the link and play right in the browser.

The "Official Account to RedNote Image Assistant" previously built in Miaoda can also keep iterating inside DuMate — accumulated skills are reusable across scenarios.

Verifying the Results

After running through the two scenarios above, you should be able to confirm these things:

• The cross-app pipeline works: the agent pulls data from Feishu and writes local files — a closed loop.
• Skills are reusable: the mini game generated by the Miaoda skill deploys straight to a live URL.
• Observability: task progress, files, and invoked capabilities are all visible in real time across three panes; debugging is never blind.

FAQ

• Q: Will the agent delete files recklessly? A: It has full control inside the authorized workspace, but delete operations trigger a second confirmation; keep the workspace confined to a dedicated directory.

• Q: Are free-tier credits enough? A: Enough for everyday office tasks; heavy agent tasks (like long code runs) consume noticeably more credits. DuMate's Harness framework claims to cut credit consumption for similar tasks by 75%, but actual usage still depends on task complexity.

• Q: How does it compare with Codex / Claude Code? A: DuMate targets office users who don't write code, with cross-app and scheduled tasks as the headline; Codex / Claude Code target developers and code engineering. The positioning differs — DuMate is one low-barrier option for users in China to experience agents.

• Q: Which app connections are supported? A: Currently Chrome, Feishu, DingTalk, GitHub, Trip.com, Baidu Netdisk, Baidu Miaoda, and more; the list keeps growing.