Skip to content

Library and Memory: Let WorkBuddy remember your materials and preferences ​

The first 3 lessons covered "how to do a single task well"; this lesson covers "the parts repeated in every task." WorkBuddy Library is the official home for output—a shared directory tree where files from both people and the Agent converge, so tasks can pull materials in and keep working on them at any time, and finished pages can even become links others can open. Memory, meanwhile, extracts preferences and habits from your conversations every night, and future conversations reference them automatically. With materials in the Library and preferences in Memory, the skills from earlier lessons—stating requirements, experts, scheduled tasks—all get smoother the more you use them. In 12 minutes, hand your materials and preferences over to WorkBuddy to remember.

This is lesson 4 of "Efficient Workflow" (the final lesson in this series), based on WorkBuddy v5.9.3, with content drawn from the official documentation for "Library," "Memory," and "My Mailbox."

What you'll be able to do after this lesson ​

  • Store files in the WorkBuddy Library, knowing what goes in "My Documents" versus "Team Spaces"
  • "Add" spaces or documents from the Library "to a task," let the Agent work based on existing materials, and save the results back to the Library
  • Connect Tencent Docs and the Lexiang knowledge base, then search, reference, and write back in one place after authorization
  • View, edit, or delete what WorkBuddy automatically remembers under "Settings → Memory," or import memories from other AIs
  • Enable your dedicated WorkBuddy mailbox so incoming mail turns directly into work for the Agent

Your current struggles ​

  • Re-uploading all your project materials and re-explaining the project background every time you start a new task
  • The proposal you had AI write last week—you want to keep editing it this week but can't find it in your chat history
  • Preferences like "I like reporting with tables" or "use formal titles with clients" have to be restated in every new conversation

When to use this technique ​

When you need to:

  • Have the Agent work based on existing materials—put materials in the Library first, then use them in tasks anytime without repeatedly explaining the project
  • Build up the output you created together with the Agent, so you can keep editing and adding to it next time
  • Let WorkBuddy remember your preferences and habits—in the official words: "The more you chat, the better it knows you"

And you don't want to:

  • Dig through dozens of conversations to find that file generated last week
  • Drag the same PDF into the chat box again for every new task
  • Explain your working habits from scratch every single time

🎒 Preparation before you start ​

  • Read 3.0 Efficient Workflow: From Knowing How to Using It Well and know the roadmap for this series' 4 lessons
  • Know how to create tasks and send messages—if you're not fluent yet, revisit 2.3 A Complete Guide to Input Methods
  • WorkBuddy client is logged in—this lesson's steps follow the desktop client interface
  • Keep one or two frequently used materials handy (project docs, weekly reports, spreadsheets all work), and store them into the Library as you follow along
  • 12 minutes of reading time

Core idea ​

WorkBuddy uses three places to remember you, each with its own job: the Library stores files, Memory stores preferences, and My Mailbox receives incoming mail.

CapabilityWhat it remembersHow to use it
LibraryFiles and output: MD / CSV / HTML, PDF / Word / PPT / Excel, images, audio, and videoAdd to a task so the Agent can work based on the materials
MemoryPreferences and habits: automatically extracted from conversations and organized every nightAutomatically referenced in future conversations—no need to repeat yourself
My MailboxExternal mail: a dedicated mailbox addressAdd emails to a conversation for the Agent to handle

How the Library differs from an ordinary cloud drive: in the official words—it doesn't just pile files together; it lets the three carriers MD, CSV, and HTML reference and interact with each other: documents are referenced by pages, spreadsheets are rendered by pages, and a change in one place takes effect everywhere. Each of the three carriers has its own role:

FormRolePositioningTypical uses
MDDetermines contentNarrativeProposals, weekly reports, documentation
CSVStores dataSingle source of truthLists, schedules, inventory, performance
HTMLHandles presentationDisplayReads CSVs and renders them into charts and lists

To change content, edit the MD; to change data, edit the CSV; to change appearance, edit the HTML—each without interfering with the others.

At the heart of the library is a working loop: add materials from the library to a task → ask questions based on the materials so the Agent can execute → save the results back to the library. Materials are not uploaded once and forgotten; they enter this continuously turning loop—the next time you analyze, the Agent directly reads the accumulated materials and continues from existing conclusions, without starting from scratch or digging through historical sessions.

There is also one principle that runs throughout: the Agent has the same permissions as the user. When the Agent reads or modifies library content in a task, its permissions are exactly the same as the current user's, with no overreach—content the user cannot see, the Agent cannot read either.

Follow Along ​

Step 1: Open the library and get to know "My Documents" and "Team Spaces" ​

Why: Know where things go before putting them there. The library has two areas, one for personal output and one for team materials—put things in the right place, and they will be easier to find, share, and authorize later.

Steps:

  1. Open the WorkBuddy client and click "Library" in the left sidebar.
  2. Get to know the two areas: My Documents (personal space, the default destination for what you and the Agent produce together, visible only to you) and Team Spaces (project materials shared with members of the space).
  3. Take a look at the capacity bar at the bottom of the library sidebar to see the remaining space.

You should see: In "My Documents" you can create directories yourself and manually create and organize content; the Team Spaces section lists the teams you belong to, with a + New Space button to the right of the heading.

💡 Find materials without entering the library

Click the search button in the top-left corner of the window to open global search. The "Library" category supports searching by the titles and body content of library objects, with keyword hits highlighted; clicking a result takes you directly to the library preview page. See 2.2 Task Management for details.

  • I can state what goes in each of the two areas: My Documents holds personal output, Team Spaces holds shared project materials

Step 2: Store materials in the WorkBuddy library ​

Why: Materials must be in the library before anything else can happen—task references, retrieval and reuse, and team sharing all depend on "storing them first." Based on the form of the materials at hand, choose one way to store them:

Option A (recommended): Have the Agent save it in a session ​

Best for MD documents and HTML pages written together with the Agent. In the session where the artifact was generated, simply ask for it to be saved:

text
Save this proposal document to the library

If there are multiple artifacts in a session, you can specify which one to save. The Agent will complete the upload through its library capability and tell you where it is stored—the artifact goes straight from "content in a session" to "a manageable document in the library," with no manual export and re-upload needed.

Option B: Upload local files and clip web pages ​

Best for existing office files and assets:

  • Office and local files: PDF, Word, PPT, Excel
  • Images, audio, and video: PNG, JPG, MP3, MOV
  • Archives: ZIP files are saved as ordinary files; when an archive contains web page content (such as index.html and accompanying static assets), the details page will prompt you with the option to "convert to an online web page"
  • Clipped web links: add articles or web page links into the directory tree

Option C: Sharing is storing ​

Best for content that is finished and ready to be sent to others (web pages, reports, and the like). Click Share on the artifact, and in the Publish tab enable Publish as Website—the act of publishing itself completes the storing, so sending it out and saving it happen in one step (see Advanced Tips for details).

You should see: The file appears in the library directory tree, where you can drag, move, and nest it into categories at any time.

💡 Have the Agent generate directly into the library

You can also specify in a task that an artifact be generated into a certain directory or team space, with no second round of moving needed.

  • I stored at least one item into the library using at least one of these methods

Step 3: "Add materials to a task" so the Agent can work based on them ​

Why: This is the core use of the library. Once materials enter the task context, in the official words: "Because the background is already in the context, there is no need to repeatedly explain the project situation." Choose either of the two entry points:

  1. File details page: Click a document to view its details, click ··· in the top-right corner of the details page to open the more menu, and choose Add to Task.
  2. Library sidebar: Select a space or document in the directory, click ··· to open the menu, and choose Add to Task.

Then ask questions directly based on the materials in the input box, for example:

text
Based on this material: compile all unfinished items, group them by owner, and output a list

For how to write clearer prompts, revisit the three-element formula in 3.1 Request Methodology. After the task is done, save the new analysis results back to the library—this loop lets historical results be reused repeatedly by later tasks.

You should see: The selected space or document enters the task context, and the Agent works by directly citing the material's content instead of asking about the project background.

💡 Prefer spaces over individual files

Official usage advice: when adding to a task, prefer selecting a directory or space over picking files one by one—the Agent gets a more complete context.

⚠️ Check for sensitive information before adding

Before adding a space or document to a task, confirm it contains no sensitive information that should not enter the conversation context (official security reminder).

  • I added a material to a task, and the Agent worked from it without asking about the background

Step 4: Connect Tencent Docs and Lexiang Knowledge Base ​

Why: Team materials are often not local—online spreadsheets live in the company's Tencent Docs, and policies and regulations live in the Lexiang Knowledge Base (Tencent Lexiang). The official name for this kind of capability is "knowledge base": authorize the connection in the library, and afterwards searching, referencing, and writing back are all done in one place, with no download-and-reupload needed.

Tencent Docs:

  1. Click Library in the left sidebar → select Tencent Docs → click Go Authorize Now (WeChat / QQ QR-code login supported).
  2. Authorization scope: read document content, edit and manage files.
  3. After authorization you can: semantic search (create a new task and describe the document you need in natural language to find it), keyword search (the search box in the top-right corner of the library, supporting search by title and by owner), and check files to Add to Task—local files that Tencent Docs can open directly, such as Word, Excel, PPT, and PDF, can also be checked.
  4. Saving artifacts back: in the artifacts panel on the right, click Upload to Cloud → select Tencent Docs; when uploading a Markdown (md) artifact, it is created as a smart document by default, with content and formatting consistent with the original md.

Lexiang Knowledge Base:

  1. Library → select Lexiang Knowledge Base → Go Authorize Now (WeChat QR-code / phone number login; the first login requires binding a phone number).
  2. Authorization scope: read the knowledge base, search and query knowledge content, create and edit knowledge content.
  3. Two ways to reference in tasks: Add to Task from within the Lexiang Knowledge Base; or open the Lexiang Knowledge Base from the upload file icon above the new-task input box and check an entire team space, knowledge base, or file.
  4. Saving artifacts back: Upload to Cloud → select Lexiang Knowledge Base, with storage by team space supported.

You should see: Once authorization is complete, Tencent Docs / Lexiang Knowledge Base entries appear in the library; the Tencent Docs file list also labels each document's ownership (personal / external), making it easy to tell where a document comes from—check it and you can add it to a task. To rebind or disconnect: click the corresponding knowledge base, then More (⋯) in the top-right corner → Unbind (for Tencent Docs, in-app unbinding is supported only for personal accounts).

💡 There is also the ima knowledge base

The "Upload to cloud" output option also supports ima knowledge bases—once stored, content is retained and searchable as a knowledge asset, and authorization is likewise required; if ima or Lexiang does not support the current file type, the corresponding option is grayed out. For hands-on examples with ready-made Prompts for Tencent Docs, see C3 Tencent Docs Connector; for the connector's full capabilities, see 5.2 Connectors and MCP.

  • I have connected at least one knowledge base and referenced a file from it in a task

Step 5: View and manage WorkBuddy Memory ​

Why: The memory feature is enabled by default—WorkBuddy organizes each day's conversations every night, extracting personally relevant information of reference value from conversation history (facts, preferences and habits, relationships, recent follow-up items) to form "memory data"; in subsequent conversations or task executions, relevant memories are referenced as background information. What you need to do is know what it records and how to manage it.

Path: Click avatar → Settings → Memory.

OperationHow to do it
ViewMemory summaries are regenerated every night; click a generated memory to view it
EditClick the edit icon in the upper-right corner and tell WorkBuddy in a conversational way what to remember or forget
DeleteClick the delete icon in the upper-right corner to clear the current memory
Turn offTurn the feature off with one click; once off, extraction of new memories stops
ImportClick Start Import: copy the sample prompt into a conversation with another AI → paste the result back → click Add to Memory

Memory also supports retrieval of conversation history—when you ask "what did I do last week" or "the XX we talked about before," relevant memories are referenced as background information.

You should see: The memory list contains preferences and items mentioned in your recent conversations, visible only to you.

💡 Memory does not consume credits

Memory extraction is performed at WorkBuddy's own expense and does not additionally consume your tokens / credits.

  • I have viewed my own memory contents under Settings → Memory, and I know how to edit it and turn it off with one click

Step 6: Enable "My Mailbox" to give the Agent a work inbox ​

Why: WorkBuddy "My Mailbox" creates a dedicated email address for each user (in the form xxx.agent@agent.qq.com), which is an independent assistant identity and does not send via your personal email. The official one-line framing: Email is the Agent's "work inbox"—read incoming mail in the interface, do the work in the conversation.

Steps:

  1. Entry point: In More on the left sidebar, "My Mailbox" is located below "My Files."
  2. First-time use requires manual activation: just agree to the supplementary agreement (accounts registered with a phone number need one extra step of an SMS verification code). You can also activate it via the activation card that pops up the first time you ask the AI to send an email in a conversation, without interrupting the current task.
  3. Manage email in the interface: Inbox / Sent tabs, with emails grouped by Today, Yesterday, Last Week, and Earlier; supports marking as read, deleting, and searching by keyword; attachments show file size and can be downloaded.
  4. Have the Agent handle email: Click Add to Conversation in the email list or detail page, and the entire email (subject, body, sender, time, attachments) is passed in as context for WorkBuddy to summarize, extract information, or draft a reply.
  5. Writing emails happens in the conversation: have WorkBuddy draft it, and it is sent only after you confirm—outbound actions like replies and forwards are not sent automatically; if you send without adding anything, WorkBuddy will first ask how you want to reply and to whom.

You should see: New emails appear in the notification center at the top of the left sidebar; once expanded, you can "Add email to conversation" or "Go to mailbox" with one click.

💡 Viewing email does not consume credits

Viewing and managing email is done in the in-app interface and does not consume tokens; only when you have WorkBuddy process email or write email in a conversation are credits consumed as a normal task (see 2.6 Credits and Pricing).

  • I understand the mailbox's division of labor: read incoming mail in the interface (no credit cost), do the work in the conversation (credits consumed per task)

For details, refer to the official documentation: WorkBuddy Library Feature Description, Memory Feature Description, My Mailbox Feature Description.

Checkpoint ✅ ​

  • Can state the division of labor between the two areas of the library: My Documents holds personal output (visible only to yourself), while Team Spaces hold project-shared materials (visible to members)
  • Saved a document to the library, and in a new task used "Add to Task" so the Agent could work based on it
  • Connected either a Tencent Docs or Lexiang knowledge base and can reference files from it in tasks
  • Have viewed my own memory under Settings → Memory, and know that it regenerates every night, is visible only to me, and does not consume credits
  • Can distinguish which operations do not consume credits: storage, directory organization, retrieval, manual editing, viewing email, memory extraction—only AI processing, analysis, and revision consume credits

Pitfall Reminders ​

❌ Leaving outputs only in the conversation means they can't be recovered after the conversation expires ​

Symptom: A proposal you wrote with the Agent last week—you want to continue editing it this week, but you can't find the file anywhere in your conversation history.

Cause: The output was not saved to the library, so it cannot be recovered after the conversation expires—in the official usage advice's own words: "Promptly save important outputs back to the library to avoid being unable to recover them after the conversation expires."

Correct approach: When a task is done, have the Agent save key results back to the library as a matter of habit; or click Share on the output and enable Publish as Website—sharing and saving in one step.

❌ Adding an entire Team Space to a task sends all sensitive information into the context ​

Symptom: You only wanted the Agent to see the schedule, but after adding the entire space to the task, contracts and compensation data in the space also entered the conversation context.

Cause: What was added to the task was the entire space, and the Agent will read everything in it that you have access to.

Correct approach: Official security reminder—before adding a space or document to a task, confirm it contains no sensitive information unsuitable for the conversation context; add only the documents you need, or move sensitive files out first.

❌ Treating memory as 100% reliable fact ​

Symptom: WorkBuddy mentions some piece of information as "something you said last time," but it differs from what you actually said, or is even outdated.

Cause: Memories are automatically extracted by the model from conversation content; the official documentation explicitly states they "may contain summarization bias or timeliness issues."

Correct approach: Verify before relying on memory for important decisions; for incorrect entries, edit, correct, or delete them under Settings → Memory; if you don't want certain information remembered, delete the corresponding entries or turn the feature off with one click. Note: After turning it off or deleting, previously generated outputs will not be retroactively retracted.

❌ Leaving "Publish as Website" on and forgetting to turn it off ​

Symptom: A link sent to a colleague can still be opened months later, and the outdated data referenced on the page remains exposed.

Cause: Once "Publish as Website" is enabled, anyone who obtains the link can access it, and it does not turn off automatically; published links use the workbuddy.link domain, and page content is the publisher's responsibility.

Correct approach: Before publishing, confirm the page and the data it references are suitable for external access; when no longer needed, return to the share dialog and toggle off "Publish as Website" to stop access.

Advanced Tips (Optional) ​

🔧 Lightweight publishing: turn local HTML into an online link with one click. Local HTML files sent to colleagues often won't open—a local path only points to your own computer. Click the share icon in the upper-right corner of the page → Publish tab → turn on Publish as Website, and you get an online link on the workbuddy.link domain (in the form https://workbuddy.link/p/{页面 ID}). Going further, HTML can pull in a CSV as a data source: data stored online with two-way sync—for lightweight apps like project management boards or customer follow-up sheets, one HTML page plus one CSV table is enough.

🔧 Team Space collaboration: four permission tiers + AI revisions that don't overwrite the original text. Space member permissions come in four tiers: view, edit, manage, and no permission; the Agent reads according to the current user's permissions and cannot read what a person cannot see. Team-shared MD documents are discussed via inline text-selection comments, and AI modifications do not overwrite the original text—each change is presented as a revision suggestion pending confirmation and is written into the body only after acceptance; when you receive a batch of comments, use "Revise by comments" to have the AI produce suggestions one by one. Credit rules: whoever clicks and sends the AI instruction consumes their own credits; accepting or rejecting revision suggestions itself consumes nothing.

🔧 Data management: a unified hub for sharing and archiving. Under avatar → Settings → Data Management, "Files I shared" lets you copy links, download, and cancel sharing (after cancellation, others can no longer access via the link); "Archived tasks" can be deleted (with double confirmation; after deletion they cannot be viewed or continued as conversations) or unarchived (restored to the task list).

🔧 What to do when your storage is full. Before an upload, the system first checks the remaining space; when capacity is full, a popup shows the current version's limit: you can clean up files to free up space, or click to upgrade for more capacity—Trial 5GB, Standard 10GB, Advanced 20GB, Flagship 50GB, and the quota takes effect in real time after upgrading. Also, a note on where things are stored: uploaded office and local files are stored by Tencent Cloud Drive, while content created in WorkBuddy is stored in WorkBuddy cloud.

Lesson Summary ​

✅ WorkBuddy Library = a shared home for outputs from both you and the Agent: My Documents holds personal output, Team Spaces hold project materials, and a single directory tree contains MD / CSV / HTML, office files, and images, audio, and video.

✅ Core loop: add to a task → ask questions based on the materials → save results back to the library; the Agent has the same permissions as you and cannot access beyond them.

✅ Knowledge base connections: Tencent Docs (after authorization, semantic / keyword search, citation, and write-back; md is saved as a smart document by default), Lexiang knowledge base (stored by team space), and ima knowledge base (for accumulating knowledge assets).

✅ Memory regenerates every night: view, edit, delete, or turn it off with one click under Settings → Memory; supports importing from other AI; visible only to you and consumes no credits.

✅ My Mailbox = the Agent's work inbox: read incoming mail in the interface (no tokens consumed), do the work in the conversation (credits consumed per task); outgoing messages must be confirmed by you.

Key facts recap:

ItemKey pointDescription
Library entryLeft sidebar → LibraryMy Documents + Team Spaces
Library capacityTrial 5GB / Standard 10GB / Advanced 20GB / Flagship 50GBCheck the capacity bar at the bottom of the sidebar
Memory entryAvatar → Settings → MemoryRegenerates every night, visible only to you
Memory contentFactual information, preferences and habits, relationships between people, recent follow-upsExtracted automatically by the model; verify before important judgments
Mailbox entryLeft sidebar, More → My MailboxDedicated address like xxx.agent@agent.qq.com, requires manual activation
Operations that consume no creditsStorage, directory organization, retrieval, manual editing, reading email, memory extractionOnly AI processing, analysis, and revision consume credits; whoever issues the instruction is charged

Next Lesson Preview ​

With this, the "Efficient Workflow" 4-lesson series is complete: stating requirements clearly, experts improving quality, automation taking over repetition, and the library accumulating background. The next lesson moves into practice—4.0 Scenario Practice: Choose Your Route.

You will learn:

  • How the Scenario Practice section helps you pick a route based on your work: weekly reports, meeting minutes, Word proposals, PPT, Excel, Tencent Docs, WeCom remote work, and Skills practice are all included
  • The complete workflow for each scenario, with the original official Prompts ready to copy
  • How the methods from the previous 4 lessons apply to concrete pieces of work

Related Tutorials ​

gaodong.aiAboutPrivacyContact

Independent tutorials, not affiliated with any product mentioned. Product names and trademarks belong to their owners. Check official pages for prices and policies.