Table-based conversational search lets you run natural-language queries against structured data in tables. After you define a simple table schema, your end users can ask questions like:
"Which product had the highest sales in Q3?"
"Show me all orders placed between January and March with a value greater than $500."
"What is the return rate for product SKU-1023?"
OpenSearch LLM-Based Conversational Search Edition maps these natural-language queries to the underlying table, retrieves matching rows or computed answers, and returns a structured response — without requiring you to write SQL.
How it works
Create a secondary table in your OpenSearch instance and define the field schema.
Load data into the secondary table — by importing a file or syncing from a MaxCompute data source.
Run a conversational search test to verify results.
During a search, OpenSearch queries the secondary table first. If Text Q&A is enabled and no answer is found in the table, the search falls back to the field tagged as Content, which increases the overall response rate. Note that a field with the Content tag is synchronized to the main table.
Prerequisites
Before you begin, make sure you have:
An OpenSearch LLM-Based Conversational Search Edition instance
(For MaxCompute sync) An AccessKey ID and AccessKey Secret for the account that owns the MaxCompute project. To create one, see Create an AccessKey pair
Limitations
Each instance supports up to 5 secondary tables.
Each secondary table supports up to 30 fields for writes and queries.
Field schema constraints
Review these constraints before configuring your table schema:
| Field property | Constraint |
|---|---|
| Table name | Letters, digits, and underscores (_); 1–20 characters |
| Field name | Must start with a letter; letters, digits, and underscores (_); up to 30 characters |
| Field description | Up to 30 characters; strongly affects search quality — specify a description for every field |
| Primary key type | INT or TEXT; each value must be unique |
| Field type | Must match the type in the source table; mismatched types cause upload failure |
| Content tag | Field must be TEXT type; only one field per secondary table can carry this tag; synchronized to the main table |
| Field tag (Text Q&A off) | Not required |
| Field tag (Text Q&A on) | Must be set to Content or Primary Key |
Choose a setup method
| Method | Best for |
|---|---|
| Manually create a secondary table | Small datasets, one-time uploads, or when no persistent data source exists |
| Create a secondary table via a data source | Ongoing sync from MaxCompute; the table schema is derived directly from the MaxCompute table |
Method 1: Manually create a secondary table
Create the secondary table
Go to the Instance Management page, find your instance, and click Manage in the Actions column.
In the left-side navigation pane, choose Configuration Center > Data Configuration, then click Change Configurations.
In the Data Processing Configuration step, keep the default settings and click Next to reach the Data Structure step.
Click Add Table and select Manually Create.
Define your fields. To enable text-based fallback search on a specific field, turn on Text Q&A and assign the Content tag to that field.
Click Completed and wait for the secondary table to be created.
Load data
Choose one of the following methods:
Option A: import a file
Click Import File. Review the sample data format before importing.
To upload data programmatically, use the API instead. See Push documents.
Option B: sync from MaxCompute
Click Add Data Source and select MaxCompute.
Click Connect to Database. In the dialog box, enter the Project Name, AccessKey ID, and AccessKey Secret, then click Connect.
To reduce the threat of disclosure, the AccessKey Secret is displayed only once upon creation and cannot be retrieved. Store it securely before proceeding.
Select the MaxCompute table to use and click OK.
Map MaxCompute fields to the fields in your secondary table, then click OK.
Specify partition filters if needed, then click Completed. If no filters are specified, data is pulled from all partitions.
Method 2: Create a secondary table via a data source
Use this method when your structured data already lives in MaxCompute and you want the field schema derived automatically.
Go to the Instance Management page, find your instance, and click Manage in the Actions column.
In the left-side navigation pane, choose Configuration Center > Data Configuration, then click Change Configurations.
In the Data Processing Configuration step, keep the default settings and click Next to reach the Data Structure step.
Click Add Table and select Use Data Source.
In the Select Data Source panel, click MaxCompute, then click Connect to Database. Enter the Project Name, AccessKey ID, and AccessKey Secret, then click Connect.
Select the MaxCompute table to use and click OK.
To enable text-based fallback search on a specific field, turn on Text Q&A and assign the Content and Primary Key tags to the appropriate fields.
For details on other field tag options, see Import data.
Specify partition filters if needed, then click Completed. If no filters are specified, data is pulled from all partitions. Wait for the MaxCompute data source to connect and the secondary table to be created.
Test the configuration
After data is loaded, run a conversational search test to verify results.
Go to the Perform a Q&A test page and enter a natural-language question based on your table data.
Check the response. A correct response returns a structured answer drawn from the secondary table. If Text Q&A is enabled and no table match is found, the response falls back to the Content-tagged field.
To see the total number of documents indexed, go to Data Configuration and check the Data Query section.
To query data programmatically, see Q&A document query.
Next steps
Tune field descriptions to improve search accuracy — clear, specific descriptions produce better results.
To update data, re-import the file or add the MaxCompute data source again to sync data.
To add more tables, repeat the steps above. Each instance supports up to 5 secondary tables.