Multi-source, multi-engine DataAgent

Create a connection

1. Click Data Source Connections in the left-side activity bar.

2. Click + at the top of the panel.

3. Select a data source type and an authentication method.

4. Configure the connection parameters and click Test Connection.

5. Click Save and Connect.

After the connection is established, view its status and details in the Data Source Connections panel, or connect, disconnect, and delete the data source.

Configure data sources

Run SQL queries

1. Create or open an SQL file.

2. Select a connection at the top of the editor.

3. Enter an SQL statement or generate a query from the Catalog.

4. Run the entire script, the selected content, or the statement at the cursor position.

5. View data and diagnostic information in the results section.

Run Milvus retrieval

Retrieval Builder generates Milvus JSON operation scripts from a form. A collection must be selected in all modes. After selecting a collection, adjust the output fields as needed.

1. Select a Milvus connection at the top of the editor.

2. Expand Retrieval Builder and select a collection.

3. Select a retrieval mode and configure the required inputs.

4. Select the output fields for the results as needed.

5. Click Generate Script to write the JSON script to the editor, or click Run to run it directly.

Manage OSS objects

On the OSS bucket management page, browse objects in column, list, or grid view; create folders; upload, download, rename, or delete files; and view transfer progress and preview content. AI Query can list buckets and objects, read file samples, and generate MaxCompute query SQL or external table DDL.

View results and visualizations

After running an SQL query or retrieval, switch views based on the result type:

• Table and JSON: View raw rows and columns or structured responses.

• Charts: Display results in bar charts, line charts, area charts, pie charts, scatter charts, radar charts, treemaps, funnel charts, Sankey diagrams, relationship diagrams, 3D charts, metric cards, and other chart types.

• Pivot and quality analysis: Aggregate dimensions and metrics, and view null values, distributions, and anomalies.

• Vector analysis: View distance distributions, nearest neighbors, principal component analysis, and heatmaps.

• Execution diagnostics: View MaxCompute execution logs and Logview, or view Hologres, StarRocks, and Milvus job monitoring information.

After configuring a chart, pin it to a dashboard for further analysis. For more information, see Create dashboards.

Create a knowledge space

1. Open Knowledge and Memory in the left-side bar.

2. Click + and select the folder that contains the materials.

3. Confirm the space name, vector model, file types, and scheduled synchronization option.

4. Click Create and Sync.

5. Wait until the space status displays the number of synchronized sources.

After synchronization is complete, search the materials in the space by using natural language, view referenced snippets grouped by document, preview original files, and manually or periodically synchronize changed files.

Manage My Memories

Memories can be saved by AI during conversations or manually created in My Memories.

1. Open Knowledge and Memory > My Memories.

2. Click New Memory and configure the subject, relationship, content, and tags.

3. Click Save.

The memory list supports search, tag-based filtering, date-based grouping, and pinning. After selecting a memory, edit its content and tags, or delete the memory if it is no longer needed. Source labels distinguish manually added content from content written by AI.

Scenarios

• Joint analysis of business data and policy materials: Query data warehouse metrics while retrieving metric definitions and policy descriptions from business documents.

• Comparison of real-time and offline data: Compare MaxCompute offline results with Hologres and StarRocks real-time data to identify differences in definitions or data timeliness.

• Semantic retrieval and detail verification: Retrieve relevant content from Milvus or a knowledge space, and then query data tables to verify conclusions.

• Analysis of object files and data warehouse data: Read OSS file samples and perform joint analysis with cloud table data.

• Continuous reuse of business context: Use common definitions, preferences, and facts in My Memories to interpret subsequent results.

Start a multi-source analysis

1. Specify the analysis objective and time range in AI Query.

2. Specify the data engines, knowledge spaces, or OSS objects to use.

3. Review the approval card for write, delete, or file modification operations.

4. View the query results, referenced snippets, and visualizations returned by DataAgent.

5. Ask follow-up questions about metric definitions, anomaly sources, or breakdown dimensions.

Example questions:

• Explain the change in gross profit this month based on the MaxCompute sales details and the promotion rules in the knowledge space.

• Compare Hologres real-time orders with the MaxCompute daily settlement table and identify the channel with the largest difference.

• First, find similar complaints in Milvus. Then, use StarRocks to calculate the trend over the past 7 days for the corresponding causes.

• Read a sample of the sales file in OSS and aggregate the data based on the regional definitions in My Memories.