Tencent WorkBuddy Guide: AI Agent That Directly Edits Your Files

This tutorial covers Tencent's WorkBuddy AI agent, which directly modifies Word, Excel, and PPT files, explaining installation, three operation modes based on risk, prompt engineering with five key elements, four practical examples for file organization, data merging, research presentations, and scheduled reports, plus tips on skills, connectors, credit management, and verification habits.

Model Perspective
Model Perspective
Model Perspective
Tencent WorkBuddy Guide: AI Agent That Directly Edits Your Files

Tencent's WorkBuddy launched in March 2024 and reached 6.58 million monthly active users by July, ranking first among office AI agents according to QuestMobile. Recent updates include an independent file browser (October 8) allowing right-click editing of Word, Excel, or PDF files, and Enterprise WeChat integration (October 14) for single sign-on.

How WorkBuddy Differs from Chat AI

Unlike chatbots that only give advice, WorkBuddy executes tasks directly on your computer: it breaks a request into steps, plans, executes, and delivers finished files (Word, Excel, PPT, web pages) in a result pane while keeping drafts in a workspace folder. This means it can handle multi-step workflows like organizing folders, merging spreadsheets, and building presentations in one go, but mistakes have higher consequences because it actually modifies files.

WorkBuddy main interface showing task, assistant, expert, and automation entries with central input box
WorkBuddy main interface showing task, assistant, expert, and automation entries with central input box

Installation and Initial Setup

Download and login: Official site workbuddy.tencent.com (redirects to CodeBuddy). Windows 10+, macOS 12+ (separate Intel/Apple silicon builds), 统信UOS and 银河麒麟 via system app stores. Login via WeChat QR code.

Create a dedicated workspace folder: Use "Select Workspace" at bottom-left of input box. Example: D:\WorkBuddy工作区. Copy files to process into this folder. Do not grant access to entire drives or directories containing passwords or original contracts.

Keep default permissions: Two tiers: "Default" (asks before modifying files or operating outside workspace) and "Full Access". Beginners should stay on Default; extra confirmations are safer than recovering files later.

Three Modes: Choose by "Regret Cost"

Click the "+" left of the input box → Mode. Three options:

Default (Craft/Agent): Executes immediately, directly edits files.

Plan: Shows a plan first; you click "Start Execution" after review, or "Adjust Plan" if unsatisfied.

Ask: Chat only, no file operations, consumes fewest credits.

Mode menu showing Default, Plan, and Ask options
Mode menu showing Default, Plan, and Ask options

Selection guideline: Ask for concepts/brainstorming; Default for generating new files (easy to delete and retry); Plan for irreversible actions — batch moves, renames, overwrites, sending emails. Common pitfall: in Default mode, even a casual "how do I do this" triggers execution and spends credits; switch to Ask for pure questions.

Decision flowchart: choose mode based on regret cost
Decision flowchart: choose mode based on regret cost

Five Elements of a Clear Instruction

Official docs recommend specifying goal, input, output format, and constraints. Add a fifth: verification criteria .

Read "9月报名表.xlsx" in workspace, group by grade to count registrations and paid count, calculate payment rate per grade, generate a bar chart. Output a new Excel; do not modify original. Highlight rows with unclear payment status in yellow, and list them separately.

This prompt contains: goal (payment rate), input (specific file), output (new Excel + chart), constraint (don't touch original), verification (flag unclear rows yellow). The last item forces the AI to surface uncertainty instead of hallucinating.

Official example shows a single sentence chaining download, extraction, modeling, PPT creation, and report writing. For long tasks, split: first extract data, verify, then proceed to charts and reports — one early error cascades.

Official example: one sentence covering download, extraction, modeling, PPT, and report
Official example: one sentence covering download, extraction, modeling, PPT, and report

Four Hands-On Case Studies

Case 1: Organize a Messy Folder

Use Plan mode. Sort workspace files into five subfolders: Documents, Spreadsheets, Images, Archives, Others. Prefix duplicate filenames with modification date. List the move plan for my confirmation; do not delete any files.

Review the plan before execution. Explicit "do not delete" is more reliable than restoring from recycle bin.

Case 2: Merge Multiple Spreadsheets and Chart

Merge "一班.xlsx", "二班.xlsx", "三班.xlsx" into a master sheet, deduplicate by student ID, compute per-class average, max, and pass rate, generate a comparison bar chart, and write a ≤300-word summary into a Word doc. Leave originals untouched.

Table processing is a strong suit; official docs suggest using the MiniMax model for such tasks.

Case 3: Research Compilation into PPT

Research 2026 math modeling competitions for middle schoolers; compile each competition's registration date, format, and official link into an ~8-slide PPT. Attach source link for every item; write "Not found" for missing data — do not guess.

Research tasks often produce plausible but fabricated output. Requiring sources and allowing "Not found" makes verification faster. Always spot-check dates and numbers by opening links before delivery.

Case 4: Scheduled Weekly Report

Every Monday 8:00 AM, scan workspace for files added or modified last week; produce a list (filename, modification time, one-sentence summary); push to my WeChat.

Create a scheduled task with daily/weekly/monthly/one-time options. Note: the client must be running; enable "Allow running when locked" in settings if the machine sleeps.

Extending Capabilities: Skills, Experts, Connectors

Skills menu: search, add from local, manage skills
Skills menu: search, add from local, manage skills

Skills: Packaged fixed workflows (e.g., weekly report, standardized data cleaning). Invoke with one phrase. Caution: third-party skills may send data elsewhere; review source and requested permissions.

Experts: Domain-specific personas (100+ available). Complex tasks can use an "Expert Team" where a lead splits work among specialists. Cost: typically 3–5× a single expert's credits; check balance first.

Connectors: Channels to external services — Tencent Docs, QQ Mail, Tencent Meeting, Enterprise WeChat. After authorization, commands like "Save this report to Tencent Docs" work.

Connectors menu: Tencent Docs, QQ Mail, Tencent Meeting, Enterprise WeChat
Connectors menu: Tencent Docs, QQ Mail, Tencent Meeting, Enterprise WeChat

Credit Management

WorkBuddy uses credits shared with CodeBuddy. Personal pricing as of October 15, 2024:

Free tier: 500 credits/month

Standard: ¥99/month, 2,000 base credits

Pro: ¥199/month, 4,000 credits

Flagship: ¥999/month, 20,000 credits

Monthly/annual subscriptions have discounts; limited-time bonus credits may apply. Credit-saving tips: use Ask mode for questions, lighter models for simple tasks, reserve Expert Teams for genuinely complex work. Biggest saver: get the instruction right the first time — rework is the costliest.

Official personal pricing screenshot dated October 15, 2024
Official personal pricing screenshot dated October 15, 2024

Final Reminders

It makes mistakes. Verify numbers, dates, and cited sources before handing off.

Back up important files even when using Plan mode.

For students and parents: competitions like HiMCM have specific rules on generative AI use — disclose if used. Using WorkBuddy to organize materials, learn software, or check formatting is fine; the core thinking and writing must remain your own.

Ultimately, people who excel with WorkBuddy aren't necessarily the most AI-savvy; they are the ones who articulate what they want most clearly. That clarity is also the first step in modeling.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

prompt engineeringAI AgentproductivityTencentscheduled tasksAI tutorialfile automationWorkBuddy
Model Perspective
Written by

Model Perspective

Insights, knowledge, and enjoyment from a mathematical modeling researcher and educator. Hosted by Haihua Wang, a modeling instructor and author of "Clever Use of Chat for Mathematical Modeling", "Modeling: The Mathematics of Thinking", "Mathematical Modeling Practice: A Hands‑On Guide to Competitions", and co‑author of "Mathematical Modeling: Teaching Design and Cases".

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.