Context Graph (knowledge base) is an independent module in DataWorks for unified knowledge management. It automatically discovers data semantics, integrates multi-source business knowledge, and builds a unified and governable knowledge network to drive precise Agent responses. It can collect data assets from DataWorks workspaces and parse business documents that you upload, organizing scattered information into browsable, searchable enterprise knowledge. Knowledge bases created in Context Graph (knowledge base) can be used by AI Assistant and Data Agent.
Overview
Context Graph (knowledge base) is the unified knowledge management module of DataWorks. It provides a unified knowledge foundation for intelligent applications such as Data Agent and AI Assistant. It organizes information scattered across workspace metadata and business documents into a browsable, searchable, and authorization-controlled knowledge network, enabling intelligent applications to have consistent business context during Q&A, data retrieval, and analysis.
Knowledge bases are managed as independent instances. A single knowledge base can be reused by multiple intelligent applications without redundant construction in each application.
Key capabilities
A knowledge base supports the following key capabilities:
Automatically collects tables, fields, tasks, metrics, data standards, lineage, and data quality information from DataWorks workspaces.
Supplements assets with business names, descriptions, tags, usage recommendations, and risk alerts through intelligent analysis.
Classifies data assets into business domains and forms semantic data relationships by subject.
Accepts documents, spreadsheets, webpages, images, and ZIP archives to build a business document library.
Provides unified search across auto-generated knowledge and manually uploaded documents, with source tracing.
Governs knowledge base maintenance and access scope through administrators and sharing authorizations.
Keeps knowledge current through full builds, incremental builds, and scheduled builds.
A knowledge base does not modify tables, tasks, or files in the source workspace. The system reads relevant information and generates independent knowledge content.
Relationship with AI Assistant and Data Agent
The knowledge base has been separated from AI Assistant into an independent module called Context Graph (knowledge base). In the left-side navigation pane of the DataWorks console, it is at the same level as Data Agent and AI Assistant. You create and manage knowledge base instances in this module. Knowledge bases created in Context Graph (knowledge base) can be used in AI Assistant and Data Agent.
If you are an existing AI Assistant user, knowledge bases previously created in AI Assistant are displayed as local knowledge bases in the interface. Newly created AI Assistant Service instances no longer support creating local knowledge bases. To create a knowledge base, go to Context Graph (knowledge base) for unified creation and management. For more information, see Local knowledge bases (for existing AI Assistant users).
Use cases
You can use a knowledge base in the following scenarios:
Look up the business meaning of a table, field, or metric.
Understand the owner, business domain, upstream and downstream relationships, and data quality of a data asset.
Manage business terms, metrics, data standards, synonyms, and query examples in a unified manner.
Include policies, product manuals, data dictionaries, and operation guides in unified search.
Provide authorized business knowledge to intelligent applications such as Data Agent and AI Assistant.
Prerequisites
Complete the following preparations before you use a knowledge base:
DataWorks is activated, and you have entered the tenant in which you want to create a knowledge base.
To auto-generate knowledge, the current account has read permissions on the workspaces to be collected and the related data assets.
A serverless resource group and its associated project space that can run knowledge build tasks are prepared. The selected resource group has network connectivity and access permissions to the data sources or cloud services required for the build.
To upload business documents, you have prepared files in formats supported by the knowledge base.
Key concept
Concept | Description |
Knowledge base instance | An independently managed, built, and authorized collection of knowledge. A single instance can contain both auto-generated knowledge and manually uploaded knowledge. |
Personal knowledge base | A knowledge base type. By default, only the owner can use it. You can grant access to specified users, Data Agents, or AI Assistant instances through sharing. |
Enterprise knowledge base | A type of knowledge base. All members under the same DataWorks tenant can view and search the knowledge base by default. |
Local knowledge base | A knowledge base previously created in AI Assistant by existing AI Assistant users. These knowledge bases continue to be displayed in the console and support rebuild, export, and clearing operations. Newly created AI Assistant instances no longer generate local knowledge bases. |
Auto-generation | Collects metadata from selected DataWorks workspaces and organizes it into business domains, asset semantics, metrics, and terms. |
Manual upload | Uploads local business documents for the system to parse and add to the knowledge base. |
Build | The process of converting current data sources into browsable, searchable knowledge. |
Full build | Regenerates the complete knowledge from the current collection scope. Suitable for initial builds or full rebuilds. |
Incremental build | Processes only the changes since the last version. Suitable for routine updates. |
Business domain | A group of data assets classified by business subject. |
Semantic knowledge | Business names, definitions, relationships, tips, synonyms, and query examples organized from metadata and business documents. |
Create a knowledge base
Go to the knowledge base list
Log on to the DataWorks console.
In the top navigation bar, switch to the target region.
In the left-side navigation pane, click Context Graph (knowledge base).
The knowledge base list displays the knowledge bases accessible to the current account in card view. Each card shows the name, description, your role in the knowledge base, knowledge base type (personal or enterprise), owner, and creation time. You can search by knowledge base name in the search box at the top of the page.
Create a knowledge base
On the knowledge base list page, click Create Knowledge Base.
In the Create Knowledge Base dialog box, configure the knowledge base information.
Parameter
Description
Name
Required. Enter an easily identifiable name that reflects the business scope, such as "Retail Operations Knowledge Base".
Type
Required. Select Personal or Enterprise: A Personal knowledge base is visible to and manageable by only its creator. An Enterprise knowledge base is visible to all members under the same tenant but manageable by only its creator. For more information, see Knowledge base types and access scope.
Description
Optional. Describes the business scope, primary users, and update method of the knowledge base. A maximum of 1,024 characters are supported.
Click Create Knowledge Base.
After creation, you can add content to the knowledge base through auto-generation or manual upload. On the knowledge base card in the list, you can also perform Share, Administrator Settings, Edit, and Delete operations on the created knowledge base.
Knowledge production
In the knowledge production module, you can build and enrich a knowledge base in the following two ways:
Auto-generation: The system automatically scans and collects your data assets, including tables, tasks, metrics, lineage, code, and data profiles as technical metadata, and builds core semantic knowledge based on this information.
Manual upload: You can also manually upload local documents accumulated by your team, such as business glossaries and design documents, as important supplements to the knowledge base.
Auto-generation
Depending on the completeness of source assets and their metadata, results may include:
Business domains and asset distribution.
Business names and descriptions for tables, fields, and tasks.
Asset granularity, layering, owners, engines, and health information.
Field semantic roles, sensitivity types, and confidence levels.
Metrics, logical tables, business terms, and data standards.
Synonyms, query examples, and cross-table association examples.
Applicable scenarios, known risks, and analysis tips.
Table-level and field-level relationships and semantic data graphs.
Not all assets produce all of the above content. The quality of generation depends on whether the source metadata, asset naming, comments, samples, profiles, and lineage information are complete.
Full build
The first build is a full build: the system scans all assets in the selected workspaces and generates complete knowledge. After the first full build succeeds, you can use incremental builds to continuously synchronize changes. For details, see Incremental build.
Go to the target knowledge base.
Go to the Auto-generation page under knowledge generation, and click Start Building Knowledge Base.
In the Create Knowledge Production Task dialog box, configure the scan scope and the runtime environment for this task.
Parameter
Description
Scan scope
Select the DataWorks workspaces from which to collect assets.
Data profiling
Data profiling is a knowledge-base-level capability switch located in the information bar at the top of the knowledge base details page. When enabled, the system supplements data sampling and profile statistics during knowledge generation to improve semantic quality.
Data profiling triggers SQL sampling queries (MaxCompute, Spark, etc.) and incurs a small amount of compute resource consumption. Tables that fail or time out are automatically skipped without affecting the overall build. Once enabled, this setting takes effect for both subsequent full builds and incremental builds.
Runtime workspace
Select the workspace in which to run this build task.
Runtime resource group
Select the serverless resource group for running the scan and build task.
Large model
Select the large model to use for knowledge generation and semantic analysis.
NoteKnowledge scanning invokes a large language model (LLM) for semantic analysis, which consumes tokens. Initiate build tasks based on your actual needs.
Click Start Building.
View the status, progress, and logs in the task list.
After the build succeeds, go to the business domain, semantic knowledge base, or asset details page to review the results.
Incremental build
After the first full build succeeds, you can use incremental builds to continuously synchronize workspace changes. An incremental build processes only changes relative to the previous version, making it faster and less costly. Incremental builds include manual incremental builds and scheduled incremental builds. Both require a successful full build before they can be used.
Manual incremental build
Manually trigger an incremental build on demand. This is suitable for updating knowledge immediately after small-scale data changes. Procedure:
On the Auto-generation page, in the actions bar, click .
In the Confirm incremental build? dialog box, confirm the workspace scope for the incremental build, and select the Runtime Workspace, Runtime Resource Group, and Large Model.
Click Start Building.
Check the execution status of this incremental build in Run Records.
Scheduled incremental build
After you enable this feature, the system automatically runs an incremental build daily based on the current collection scope, keeping the knowledge base in sync with data sources without frequent manual triggers. Procedure:
In the information bar at the top of the knowledge base details page, turn on the Scheduled Incremental Build switch.
In the Confirm Enable Scheduled Incremental Build? dialog box, select a Runtime Workspace, Runtime Resource Group, and Large Model.
Click Enable. After the feature is enabled, the system automatically runs one incremental build per day based on the current collection scope.
To stop the daily automatic incremental build, turn off the Scheduled Incremental Build switch. Existing knowledge base content is not affected.
Build strategy selection
The following table lists the recommended build combinations for different stages:
Stage | Recommended combination | Description |
Initial onboarding | Full build + data profiling disabled | Quickly establishes a baseline and validates the collection scope. |
Semantic tuning after result verification | Full build + data profiling enabled | Improves semantic quality within a stable scope. |
Routine sync | Incremental build + scheduled incremental build | The scheduled incremental build automatically picks up changes without frequent manual triggers. |
Collection scope expansion | Full build + data profiling disabled | Rebuilds the baseline when the scope changes significantly, then reverts to routine incremental builds. |
View build tasks
A build task may go through stages such as input preparation, content parsing, semantic generation, knowledge publishing, and search index update. You can view the running status, progress, and logs in the task list.
A new version becomes the current browsable and searchable version only after both the build and publishing succeed. Failed or canceled tasks do not affect the last successful version.
When multiple builds run concurrently, the system prevents an older task from overwriting a newer successful result.
If a build fails, check the logs for file format issues, resource group status, workspace permissions, network connectivity, and source data access. Fix the issues and rerun the build.
If a task is in the failed state, refer to the following table to locate and resolve the issue based on log clues:
Error category | Typical log clue | Action |
File format | The log indicates that | Refer to the "Supported file formats" and "ZIP archive processing rules" sections, convert files to a supported format, and upload them again. |
Resource group | The log indicates that the serverless resource group is unavailable, has exceeded its limit, or is not associated with the workspace | Check the resource group status and its associated workspace. Refer to the "Prerequisites" section to verify the resource group configuration. |
Workspace permissions | An error indicates that access to metadata, data assets, or workspace read operations is denied. | Contact the workspace administrator to grant read permissions on the target workspace and data assets. |
Network connectivity | An error indicates a timeout or connection failure when accessing a data source or cloud service. | Check whether the resource group can reach the required data sources and cloud services over the network. |
Source data access | An error indicates that the source table or source asset does not exist or returns empty results. | Verify that the source data still exists and that the path and credentials are valid. If necessary, narrow the collection scope and try again. |
After you fix the issue, run the knowledge build task again.
Manual upload
In addition to auto-generated semantic knowledge, you can upload local business documents to supplement the knowledge base with business knowledge that cannot be collected by the system, such as metric definitions, proprietary term explanations, and policies. Uploaded documents are stored and searched together with auto-generated knowledge.
Supported file formats and limits
File category | Supported formats |
Text documents |
|
Office documents |
|
Webpages |
|
Images |
|
Archives |
|
Upload limits: You can upload up to 100 files at a time. Each regular file cannot exceed 10 MB, and each ZIP/RAR archive cannot exceed 100 MB. Legacy .doc and .xls formats are not supported. Actual limits are subject to the prompts displayed in the console.
Upload and build documents
Go to the target knowledge base and click Manual Upload in the left-side navigation pane.
In the User-defined Documents dialog box, select the runtime workspace, runtime resource group, and large language model.
Click or drag files to the upload area, and then click Upload and Build. After the upload is complete, the system automatically rebuilds the knowledge base.
View the build status on the Knowledge page. After the build succeeds, you can verify the content through knowledge search.
Uploading a file does not mean it is immediately searchable. The file takes effect only after a successful build and publish. If you replace or delete a file, you must rebuild the knowledge base for the changes to take effect.
If the uploaded files contain images, the platform relies on multimodal capabilities of large language models during the knowledge build. The models that currently support multimodal capabilities are qwen3.8-max and qwen3.7-plus.
Improve document parsing quality
Use clear headings and sections. Make sure that scanned PDFs and images are legible.
Use clear and unique column names in the first row of Excel files. Split files by topic if the content is extensive.
Remove passwords, keys, personal privacy data, and information that you are not authorized to disclose before uploading.
Use the knowledge base
View knowledge base content
After a successful build, you can view the results from the following pages:
Page | Description |
Business domain overview | View the number of business domains, table and task scale, asset layering, semantic coverage, sensitive data overview, popular tags, and collection scope. |
Business domain details | View the tables, tasks, metrics, logical tables, owners, risks, and analysis tips in a specified business domain. |
Semantic knowledge | View business terms, metrics, logical tables, cross-table query examples, field semantics, query examples, data standards, and synonyms. |
Asset details | View the business definition, field descriptions, semantic sources, confidence levels, usage recommendations, sensitive data, quality rules, lineage, and related knowledge of a single asset. |
Semantic data graph | View the relationships and upstream and downstream impacts between assets and knowledge objects. |
The content on these pages may come from source systems, DataWorks platform metadata, or system inference. The system preserves semantic sources and confidence levels whenever possible. Important business definitions and low-confidence content should be verified with the source assets and business owners.
Search knowledge
Knowledge search retrieves both auto-generated semantic knowledge and manually uploaded documents simultaneously. You can search by business terms, technical names, file names, or natural language questions. For example:
What does transaction amount meanTables related to user profilesFields containing phone numbersData quality inspection processDaily aggregation query example for order tables
When you use search results, note the following:
Evaluate whether results are applicable based on the knowledge base, business domain, object type, and source.
Click a result to view asset details or original documents and verify important conclusions.
The system does not modify or translate physical table names, field names, SQL statements, code, or IDs.
You can only search knowledge bases that you are authorized to access.
Use in AI Assistant and Data Agent
A knowledge base that is successfully created and built in Context Graph can be used in AI Assistant and Data Agent to provide business knowledge.
The knowledge bases available depend on the knowledge base type and authorization:
Enterprise knowledge base: Available to all members under the same tenant.
Personal knowledge base: Available only to the owner by default. To make it available to Data Agent or AI Assistant, the owner or an administrator must grant authorization to the corresponding instance through sharing.
Data Agent and AI Assistant cannot serve as knowledge base administrators. They can only use a knowledge base after being authorized through sharing. For details, see Share a knowledge base.
Manage knowledge bases
Knowledge base types and access scope
Type | Default access scope |
Personal knowledge base | Available only to the owner by default. The owner or an administrator can authorize specified users, Data Agent, or AI Assistant to use the knowledge base through sharing, and can add administrators through administrator settings. |
Enterprise knowledge base | All members under the same DataWorks tenant can view and search by default. Only the owner and administrators can manage the knowledge base. |
Configure administrators
You can use Administrator Settings to add tenant users who collaboratively maintain the knowledge base. Administrators have permissions to configure, build, and upload, but only the owner can delete or clear the knowledge base.
Procedure:
In the knowledge base list, find the target knowledge base and click Administrator Settings.
In the Add Administrator section, search for the target user by name or account.
Click Add on the right side of the target user.
Added administrators are displayed in the Administrator List, which also shows the owner. To remove an administrator, click Remove on the right side of the administrator. The owner cannot be removed.
Role permissions
Operation | Owner | Administrator | Regular member of an enterprise knowledge base | Object with share authorization only |
View and search knowledge | Supported | Supported | Supported | Search only |
Modify knowledge base configurations | Supported | Supported | Not supported | Not supported |
Configure and run builds | Supported | Supported | Not supported | Not supported |
Upload and delete manual documents | Supported | Supported | Not supported | Not supported |
Administrator and sharing authorization | Supported | Supported | Not supported | Not supported |
Clear knowledge base semantic data | Supported | Not supported | Not supported | Not supported |
Delete a knowledge base instance | Supported | Not supported | Not supported | Not supported |
Authorization limits
Administrators must be users in the current tenant. Data Agent and AI Assistant services cannot be administrators. They must be authorized through sharing before they can use the knowledge base. A single knowledge base supports up to 200 administrators and 200 sharing authorizations.
An enterprise knowledge base allows same-tenant members to search by default, so you typically do not need to add authorizations for individual users. A personal knowledge base can be authorized to specified users, Data Agents, or AI Assistant services through sharing as needed.
Content language
Knowledge base content supports Simplified Chinese and English. A single knowledge base maintains only one content language.
During the first build, the system determines the content language based on the language setting in the upper-right corner of the DataWorks console.
After the language is determined, full builds, incremental builds, scheduled builds, and manual upload builds all use that language. Switching the console language afterward only changes the interface display and does not change the language of existing knowledge.
You cannot directly change the content language of an existing knowledge base. To use a different language, create a new knowledge base and initiate the first build with the target page language selected.
Physical table names, field names, SQL statements, code, IDs, raw tags, and reference evidence are not translated. Therefore, an English knowledge base may still contain Chinese text from the source data.
Usage recommendations
Create knowledge bases around stable business scopes. Avoid placing unrelated workspaces and documents in the same instance.
Assign a clear owner for each knowledge base and configure backup administrators.
When you use the feature for the first time, select a small number of representative workspaces to verify results before expanding the collection scope.
Improve auto-generation quality by enriching source table and field comments, data standards, owners, lineage, and data quality rules.
Use scheduled incremental builds to keep knowledge up to date and periodically spot-check results.
Manually verify core metrics, sensitive fields, and key queries.
Periodically clean up expired documents and test assets to reduce search noise.
Follow the principle of least privilege and promptly remove authorization when access to the knowledge base is no longer needed.
Appendix: Local knowledge base (existing AI Assistant Service users)
The knowledge base has been separated from AI Assistant Service as an independent module. For existing AI Assistant Service users, knowledge bases previously created within AI Assistant Service are retained as local knowledge bases and continue to be displayed in the console without affecting existing usage.
Supported operations for local knowledge bases: Existing local knowledge base features remain available and support the following operations: regenerate, export, and clear.
Differences from Context Graph (knowledge base)
Item
Local knowledge base (existing)
Context Graph (knowledge base)
Entry point
The Knowledge Base tab on the AI Assistant Service instance details page
Context Graph (knowledge base) in the left-side navigation pane of the console
Creation
No longer generated for newly created AI Assistant Service instances
Can be created
Ownership
Belongs to a single AI Assistant Service instance
Independent instance that can be reused by multiple intelligent applications
Knowledge base type
No type distinction
Supports personal knowledge bases and enterprise knowledge bases
Sharing method
Used only within the current instance
Can be shared with users, Data Agents, or AI Assistant Service instances through the Share feature
Differences from newly created AI Assistant Service instances
Newly created AI Assistant Service instances no longer generate local knowledge bases, which means you can no longer create knowledge bases within AI Assistant Service. To create a new knowledge base, use Context Graph (knowledge base) for unified creation and management, and then use the knowledge base in AI Assistant Service or Data Agent as needed.
FAQ
Q: Why can't I find the file through search after a successful upload?
A: Uploading and building are two separate steps. Make sure that you have initiated a manual upload build and the task has been successfully published.
Q: Why can I still find relevant content through search after deleting a file?
A: After you delete a file, you must complete another build before the related knowledge is removed from the current retrieval version.
Q: Why did some files in the ZIP archive not generate knowledge?
A: Check the file format, size, and integrity of the files in the archive, and review the build log for any skip messages. Unsupported file types do not generate knowledge.
Q: What should I do if the auto-generated names or descriptions are inaccurate?
A: Improve the names, comments, owners, standards, and lineage of the source assets first, and then rebuild. System-inferred results are intended only as comprehension aids. Important business definitions should be confirmed by the responsible business owner.
Q: After switching to the English interface, why is the knowledge still in Chinese?
A: The language of the knowledge content is determined during the first build. Switching the console language does not translate or change existing knowledge.
Q: Why do some pages have no data?
A: The system generates corresponding knowledge only when source data exists and is successfully analyzed. For example, quality-related rules are not displayed if no quality rules exist, and cross-table query examples may not be generated if no reliable relationships are found. An empty page does not necessarily indicate a build failure.
Q: What is the difference between an administrator and share authorization?
A: Administrators can configure, upload content to, and build knowledge bases. Users, Data Agents, or AI Assistant Service instances authorized through the Share feature can only use knowledge for search or Q&A and cannot modify the knowledge base.
Q: Can a personal knowledge base be shared with others?
A: Yes. A personal knowledge base is available only to its owner by default. The owner or an administrator can use Share to grant access to specified users, Data Agents, or AI Assistant services.
Q: Why can't I create a knowledge base in AI Assistant?
A: Knowledge bases have been separated into an independent module for unified management. Newly created AI Assistant services no longer generate local knowledge bases. Create a knowledge base in Context Graph (knowledge base), and then use it in AI Assistant services and Data Agents.
Q: Can existing local knowledge bases still be used?
A: Yes. Knowledge bases that existing users previously created in AI Assistant services are retained as local knowledge bases and continue to appear in the console. You can still regenerate, export, and clear them without affecting your existing usage.