Overview
Knowledge Base is a container used by Buddy to accumulate external business knowledge. You can upload or connect documents such as business rules, caliber documents, data dictionaries, SQL examples, and report templates to the knowledge base. When answering questions, generating reports, or performing anomaly attribution, Buddy searches this knowledge on demand, making results more aligned with your business reality.
Knowledge bases are divided into two types:
Category | Source | Maintained by |
Built-in knowledge base | Table metadata, field metadata (including enum values), metric metadata, and lineage relationships | Automatically maintained by the platform, requiring no user action. |
External knowledge base | Business terms, business rules, caliber documents, specification documents, report templates, and more | Manually create and upload documents. |
This document focuses on the creation, maintenance, and use of external knowledge bases.
Use Limits
Limit | Description |
File Type | Upload supports any format. Formats that can be learned as knowledge: PDF, Markdown, TXT, Word/DOCX, Excel/CSV, JSON, HTML, SQL, images. |
Single File Size | No more than 5 GB |
Files per Upload | No more than 100. |
Knowledge Base Name | Up to 50 characters |
Knowledge Base Description | Up to 1024 characters |
Business Text (Data Analysis) | Up to 3000 characters |
Manually Entered Document Name | Up to 50 characters |
Manually Entered Document Length | Up to 10,000 characters |
Modes and Visibility
The visible scope and list content of the knowledge base vary with the current mode:
Current mode | Content Displayed in List | Entries |
Data engineering / data governance mode | Display only knowledge bases of the data engineering / data governance type. | Knowledge Base in the left navigation |
Data analysis mode | Display only knowledge bases and their detail pages in the current analysis space | Knowledge Base in the left navigation (after the space is entered) |
In data analysis mode, a knowledge base serves only the current analysis space and cannot be referenced across analysis spaces.
Operation Steps
Step 1: Creating a Knowledge Base
1. Select Knowledge Base in the left sidebar of Buddy.
2. Click + Create Knowledge Base in the upper-right corner.
3. In the pop-up window, enter the following information:
Knowledge Base Name (required): Use a name with business meaning, such as "GMV Metric Anomaly Detection Rules".
Description (Optional): Briefly describe the purpose.
4. Click Create. The new knowledge base will appear in the list, initially with no documents and no description.
Step 2: Adding Knowledge Documents
Open the details sidebar on the right side of the knowledge base, click the + Add Knowledge button at the bottom, and then switch between four sources at the top of the pop-up window:
Source | Scenario | Action |
Manual Entry | No existing files; directly enter or paste knowledge content. | Enter the document name and Markdown content to automatically start the learning process. |
Upload from Local | Directly upload local files. | Files are stored as volumes in the Catalog knowledge directory and automatically enter the learning process. |
Select from Catalog | Reuse files already stored in the Catalog volume directory. | Only files under file volumes can be selected. Select files individually. |
Select from Studio | Reuse development artifact files in Studio. | Only SQL, Markdown, HTML, and JSON files can be selected. |
Manual Entry
If no ready-made file is available, you can directly enter or paste the knowledge content:
1. Switch to the Manual Entry Tab.
2. Enter the Document Name (required, up to 50 characters).
3. Enter or paste the knowledge content in the Content input box. Markdown format is supported, with a maximum of 10,000 characters.
4. Click Confirm. The system automatically enters the learning and description generation process, consistent with other document types.
Manually entered knowledge documents are displayed in the list with the type md and the source Manual Entry. Operations such as viewing, re-learning, and deleting are supported.
Uploading from Local Storage
1. Switch to the Upload from Local Tab.
2. Click Select File or drag the file into the dashed box area.
3. Configure the following at the bottom of the pop-up window:
Duplicate Name Handling: Select Auto Rename or Replace Existing File.
Target Path: Select the volume directory in the Catalog to save to.
4. Click Upload. The file appears in the knowledge document list, and its status changes to Learning.
Selecting from Catalog
1. Switch to the Select from Catalog Tab.
2. In the three-level tree structure (Catalog → Schema → File), select target files by using the check boxes. Directory-level select all and partial selection are supported.
3. You can remove items one by one from the Selected (N) list on the right.
4. Click Confirm to complete the integration.
Selecting from Studio
1. Switch to the Select from Studio Tab.
2. Browse the Studio file directory and select the files you need.
3. Click Confirm. The file automatically triggers the learning process.
Tip
Knowledge files imported from Studio maintain bidirectional synchronization with their source files in Studio: when a source file is modified, the knowledge base prompts that relearning is required; when a source file is deleted, the corresponding knowledge file is also deleted and the user is notified.
Step 3: Quickly Capturing from Artifacts
Artifacts generated by Buddy in conversations (SQL, Markdown, HTML, JSON) can be saved as knowledge with one click:
1. In the upper-right corner of the conversation artifact details page, choose More > Learn as Knowledge.
2. Confirm based on the current mode:
Data analysis mode: The system directly learns the default knowledge base of the current analysis space and prompts "Successfully added to the knowledge base."
Data engineering and data governance modes: In the pop-up window, select the target knowledge bases (multiple selections supported), and after OK is clicked, a message appears indicating "Successfully added to the knowledge base."
Proactively Capturing Knowledge After Conversations
In addition to manually clicking "Learn as Knowledge", Buddy proactively identifies content worth saving during conversations and asks whether you want to add it to the knowledge base:
Trigger Strength | Trigger Conditions |
Strong Trigger (Proactive Inquiry) | Produced structured conclusions, such as caliber confirmation, attribution completion, error fix, and SOP formation. |
Medium Trigger (Weak Prompt) | You explicitly acknowledge the AI conclusion, or engage in in-depth communication on the same topic for ≥5 rounds. |
After being triggered, Buddy saves the content separately by content type:
Content type | Persistence method |
Rule | Write into the "Business Logic" global prompt, which is automatically injected with each question. |
Conclusions / Insights | Standalone knowledge document (Markdown), recalled through semantic search |
Methods / SOPs | Method-type knowledge document, recalled through semantic search |
AI organizes the conversation content into a knowledge document (automatically named and written). After the document is generated, you are prompted to review it. After confirming that it is correct, reply with "Learn" or click the learn button in the artifact to save it to the knowledge base.
Step 4: Batch Importing via Natural Language
You can create knowledge in batches directly in the dialog box using natural language:
Upload all files in demo_volume.knowledge.file to the Business Glossary for me.
If the information is incomplete when Buddy parses the instruction (for example, the target knowledge base is not specified), Buddy asks for clarification through an Ask User card. After validation passes, Buddy executes the task in batches:
Mode | Validation item |
Data engineering / data governance | Whether you have the permission to create/edit a knowledge base; whether you have the permission to access the target Catalog/Studio. |
Data analysis | Whether you have the permission to create/edit a knowledge base in the current analysis space; whether you have the permission to access the target Catalog/Studio; cross-analysis-space creation is not allowed. |
After execution is complete, Buddy returns a learning result card in the conversation, listing the target knowledge base, type, and learning status of each file.
Step 5: Maintaining Documents and Summaries
The Knowledge Document List area in the details sidebar on the right side of the knowledge base supports the following operations:
Operation | Description |
View. | Open the document preview/editing page. Jump to the source or preview online. For details, see "View Document" below. |
Reread | Single document or select multiple documents for batch operation. The button is grayed out in learning status. |
Deleting an alarm policy | Single document or select multiple documents for batch operation. Batch deletion requires confirmation. |
Regenerate summary with AI | Trigger AI summary regeneration for a single document. |
Edit summary | Switch to edit box, up to 300 characters. |
Viewing Documents
After you click View on a document, the behavior varies depending on the document source:
Source | Viewing behavior |
Manual entry | Supports preview and editing |
Local upload | Go to the preview page. Formats supported by Studio can be edited online, while other formats can only be previewed. |
Select from Catalog | Go to the corresponding data catalog. Whether download is allowed depends on the data catalog permission check. |
Select from Studio | Go to the corresponding task development directory. Whether download and edit are allowed depends on the task development permission check. |
Supported formats for online viewing of locally uploaded files:
Format | Viewing Support |
Markdown,TXT,SQL,HTML,JSON | Supports preview and editing |
PDF, images | Supports download and viewing |
Excel/CSV,Word/DOCX | Supports download and viewing |
The viewing page supports quickly switching between documents and returning, as well as saving, refreshing, and editing operations (consistent with Studio).
There are three learning statuses:
Status | Label | Action |
Learning Completed | Green checkmark | Searchable; summary available |
Learning | Blue "Learning" + spinning animation | The buttons for rereading and editing summaries are grayed out. |
Learning failed | Red "Learning failed" | Hover over it to display the failure reason. You can click Reread to retry. |
Step 6: Maintaining Knowledge Base Descriptions
In the Description area at the top of the details sidebar:
Click the AI Generate button to automatically generate a description based on the knowledge document content (takes about 2 seconds).
Click the Edit button to manually edit the content. The maximum length is 1024 characters, and you must save your changes before you can continue with other operations.
Step 7: Maintaining Business Texts (Data Analysis Mode Only)
On the knowledge base details page in the analysis space, the Business Text section is used to supplement brief business rules and implicit conventions. The system injects them into the model as a system prompt, which takes effect automatically in every conversation. For example:
- If no time range is specified in the question, data for the current year is used by default.
- For metrics involving monetary amounts, the default unit is CNY, unless the unit is specified as ten thousand CNY or USD.
- If no region is specified in the query, all regions (including overseas regions) are counted by default.
Business text supports Markdown format, with a maximum length of 3000 characters. Choose Edit > Modify > Save to update it.
Using the Knowledge Base in Conversations
Buddy uses the knowledge base in the following two ways:
Method | Description |
Automatic search | When processing business tasks, Buddy automatically determines whether to search the knowledge base (by matching relevant Skills or identifying implicit dependencies on business rules, data definitions, or SOPs in the task), retrieves relevant content from visible knowledge bases, and generates responses after integrating it with general knowledge. |
Explicit reference | Specify a knowledge file precisely in the conversation through the Agent basic capability, and answer based only on that file. |
During the thinking process, Buddy displays the tool call status for "Search Knowledge Base" and identifies the knowledge sources used in the final answer.
Description
Mode | Permission inheritance | Administrator/Editor | Viewer | No permission |
Data engineering / data governance | Workspace permissions | CRUD Operations | You can only view items. | Hide entry |
Data analysis | Analysis Space permissions | CRUD Operations | You can only view items. | Hide entry |
Verification Result
After completing the above operations, you can confirm in the following ways:
1. The knowledge base list card displays the updated number of files, number of views, and number of usages.
2. When you ask questions related to knowledge documents in a conversation, Buddy displays the "Search Knowledge Base" tool call during the thinking process and cites the source documents in the answer.
3. In the Agent Basic Capabilities pop-up, you can see the newly uploaded file under the Knowledge Files type.
FAQs
Q: What should I do if a document remains in the "Learning" status after upload?
The learning time varies depending on the document size and type. If the process takes too long or the status changes to "Learning Failed", hover over the status tag to view the failure reason and try reading it again.
Q: Will deleting a knowledge base also delete the source files in Catalog?
The source files in the Catalog will not be deleted. Deleting a knowledge document only unbinds it from the knowledge base.
Q: What should I do if a file imported from Studio is modified?
When a Studio file is modified, the knowledge base prompts that the knowledge needs to be relearned. Click Read Again to proceed.
Q: What is the difference between business texts and knowledge documents?
Business text is a lightweight supplement unique to the data analysis mode, suitable for writing implicit conventions and default values. Knowledge documents are suitable for longer, structured business rules and documentation. Both are searched simultaneously.
References
Agent Basic Capabilities Overview
Agent Basic Capabilities
Member and Permission Management