Quick start: Data Engineer

This topic is for data engineers and describes the complete process of connecting to a data source, creating an Agent, uploading business knowledge, triggering semantic processing, building intelligent responses, and finally publishing the agent for analysts to subscribe to.

Prerequisites

• You have logged in as an engineer and switched to the target workspace.

• The current account has the permission to create Agents in this workspace. If the Add Agent entry is not displayed, ask the workspace administrator to confirm the permission. Do not switch to another workspace to bypass governance boundaries.

• There is at least one available model instance in the system.

• You have prepared a demo table without sensitive information, as well as the connection address, database name, account, and credentials for the database it belongs to.

Procedure

Create a workspace

1. In the left-side workspace name drop-down list, click Create to open the Create Workspace dialog box.

2. In the Basic Information section, enter the workspace name and description (optional), and enable Public Workspace if needed. After enabling, all users can view and subscribe to Agents in this workspace.

3. In the Analyze Data Sources section, select the database type (for example, OceanBase (MySQL Mode)), and enter the host address, port, username, and password (optional).

4. After entering the connection information, click Get List:

   • First, get and select Catalog. Multiple selections are supported; when selecting multiple, databases will be distinguished by Catalog.Database.

   • Then, get and check the Database Name to analyze.

5. If you need to limit analysis usage, enter the Resource Group (optional). Resource groups must be created in OceanBase in advance. After configuration, SQL statements generated by analysis will execute under that resource group, without consuming other business resources.

6. Click Create Workspace to complete the creation.

For detailed instructions, see Create a data source.

Create an Agent

1. Open the Agent list and verify the workspace name and existing Agents to avoid duplicate creation.

2. Click Add Agent. On the Basic Information tab, enter a name and description (optional), select Publication Scope (Public or Visible within Workspace), and optionally fill in Preset Questions.

3. Switch to the Data tab and select the minimum data scope for the task. If multiple scopes exist, select the scope first, then the tables; do not grant access to all tables at once.

4. Switch to the Behavior and MCP tabs and keep the default or minimum configurations.

5. Click OK. If the submit button shows "Processing" and is temporarily disabled, stay in the current dialog until the dashboard opens or an error returns.

For detailed instructions, see Create and configure an Agent.

Upload business knowledge

On the Asset Access tab, upload domain documents, business rules, and metric definition descriptions to provide traceable business knowledge support for the Agent's Q&A and generation capabilities. The responsibilities between the semantic layer and the knowledge base are clearly defined: the semantic layer addresses "how is this metric calculated," while the knowledge base addresses "what is this business about."

1. Open the target Agent and select the Asset Access tab at the top of the dashboard.

2. In the Source File sub-tab, use the toolbar at the top to add text or upload document and spreadsheet files to enter business knowledge such as domain documents, business rules, and historical evolution.

3. Click the action menu on the right side of a file and select Add to Knowledge Base to make the file content searchable and referencable by the Agent in conversations and generation.

4. To enhance the Agent's structured capabilities, you can upload a ZIP package containing SKILL.md to the Skills section (optional).

5. Enter the retrieval evaluation from the toolbar to confirm that expected content can be retrieved.

For detailed instructions, see Knowledge base integration and user memory and Overview of knowledge bases.

Trigger semantic processing

Trigger semantic processing with one click for the integrated data sources. The system automatically identifies business objects and synthesizes calculation definitions, forming an ontological semantic model for Agent inference.

1. Open the target Agent and select the Semantic Processing tab at the top of the dashboard.

2. Click Start Processing. In the pop-up Select Processing Module dialog box, select the required modules:

   • Business Entities and Relationships: Extracts dimensions, metrics, identifiers, and infers relationships between tables.

   • Calculation Definition: Extracts metric definitions and scoring rules from indexed business knowledge and reference SQL.

3. Select the table to be processed. Business background analysis is performed on all tables in the data source; semantic extraction only processes the selected tables. Click Start Processing to confirm.

4. After processing is complete, verify the identification results in the Business Object section: whether key objects exist, if primary table mappings are correct, and if relationship directions and cardinalities are reasonable.

5. In the Calculation Metric section, check the list of calculation metrics automatically generated by the system based on the semantic model and knowledge base. Ensure the calculation logic and filter conditions of the metrics align with business requirements.

Success criteria: Both the business object and calculation metric have been generated, and key objects and metrics have been manually verified as correct.

For more information, see Semantic Processing.

Intelligent Generation

On the Intelligent Generation tab, quickly create deliverables such as reports, dashboards, data visualizations, multi-dimensional tables, workflows, spatio-temporal analytics, and document classifiers.

1. Open the target Agent and select the Intelligent Generation tab at the top of the workspace.

2. In the Build sub-tab, under Quick Create, select the type of deliverable you want to create (report, dashboard, data visualization, multi-dimensional table, workflow, spatio-temporal analytics, or document classifier).

3. Follow the prompts to fill in parameters and run the creation. Confirm that the output result meets your expectations.

4. The generated deliverable will appear in the My Builds section, where you can view and manage it.

5. Ask a simple, answerable question in the dialogue to test the deliverable's accuracy. If the accuracy does not match your expectation, return to Semantic Processing or the Knowledge Base to correct it, then ask the question again.

Success criteria: The deliverable is generated normally, and the test question's accuracy matches expectations.

For more information, see Notebook and View and Validate Analysis Results.

Publish an Agent

After completing construction, debugging, and verification, publish the agent to the Space Market. Analysts can access published build results after subscribing.

1. Click Publish in the upper-right corner.

2. After successful publication, confirm that the agent can be found in the Agent Marketplace as an analyst.

3. Handle subscription according to organizational procedures: no additional action is required if direct subscription is possible; if approval is needed, an authorized space administrator approves the subscription request.

Success criteria: Analysts can find the agent in the marketplace. After obtaining a valid subscription, they can open the workspace from My Agents and receive correct answers.

For more information, see Version Management and Visibility and Subscription Approval.