​

Tables

​

Why Use Tables?

• Query structured data - Search databases of products, customers, orders, or any tabular data
• Retrieve specific records - Find exact matches based on multiple filter criteria
• Access up-to-date information - Use dynamic data that changes frequently
• Provide precise answers - Return specific values from your database rather than general text
• Support complex queries - Enable multi-field filtering with ranges, dates, and selections
• Scale beyond text search - Handle datasets with hundreds of thousands of records efficiently

​

How Tables Work

1. Upload your data - Import CSV, JSONL, or JSON files with your records
2. Configure fields - Define data types (text, number, date, etc.) for each column
3. Set search types - Choose how each field can be searched (text search, select, range, etc.)
4. Enable for AI - Attach the table to an AI profile as a tool
5. AI queries the table - When users ask questions, the AI generates appropriate search parameters and retrieves matching records

​

Data Flow

User Question
    ↓
AI interprets question
    ↓
AI generates search parameters (filters)
    ↓
Table search executes
    ↓
Matching records returned
    ↓
AI formats results in natural language

​

Setting Up Tables

​

Creating a Table

1. Navigate to Tables in your project
2. Click Create Table
3. Enter a descriptive name
4. Click Create

​

Importing Data

• Simplest format for tabular data
• First row must contain column headers
• Best for exporting from spreadsheets

• One JSON object per line
• Preserves data types (numbers, booleans, nested objects)
• Best for database exports

• Includes both data and field configuration
• Format: {"objects": [...], "fields": [...]}
• Best for complete table definitions with search types

​

Import Process

1. Open your table
2. Click Import
3. Select your data file
4. The system automatically detects columns and loads data

​

Configuring Field Types

​

Data Types

• Text - String values (names, descriptions, IDs)
• Number - Numeric values (prices, quantities, scores)
• Date - Date only (2024-01-15)
• DateTime - Date and time with timezone
• Boolean - True/false values
• JSON - Complex nested objects
• Vector - Embedding vectors for semantic search
• Array - Lists of values

​

Search Types

• Same as data type - No filtering (display only)
• Search text - Full-text search with partial matching
• Select - Single value selection (up to 500 unique values)
• Multiselect - Multiple value selection (up to 500 unique values)

• Same as data type - No filtering
• Range filter - Min/max range selection

• Same as data type - No filtering
• Datetime filter - Date range selection

• Same as data type - No filtering
• Toggle - True/false filter

• Same as data type - No filtering (display only)

​

Field Configuration Tips

1. Click the column name to rename it
2. Use the dropdown to change data type and search type
3. Changes apply immediately

• Unique count - Number of distinct values (capped at 500+)
• Filled percentage - Percentage of non-empty rows
• Distribution - Most common values (for fields with fewer than 500 unique values)

​

Verifying Data

1. Configure all field types
2. Click Verify data
3. Review any validation errors
4. Fix issues by adjusting field types or data
5. Re-verify until validation passes

• Numbers that don’t match your number format
• Invalid date formats
• Boolean fields with non-boolean values
• JSON fields with malformed JSON

​

Format Settings

• UTC (default)
• America/New_York
• America/Los_Angeles
• Europe/London
• Europe/Berlin
• Asia/Tokyo

• US: 1,234.56 (default)
• DE: 1.234,56

​

Attaching Tables to AI

1. Go to your AI profile
2. Navigate to the Tools tab
3. Scroll to Search Tables
4. Click Add Search Table
5. Fill in the details:
   • Tool Name - How the AI refers to this tool (e.g., search_products)
   • Tool Description - When the AI should use this tool (e.g., “Search the product catalog by category, price, and availability”)
   • Select Table - Choose your configured table
6. Click Add Search Table

​

Testing Table Search

1. Open your table
2. Click Search
3. The search dialog shows:
   • Search Schema - The exact tool definition the AI sees (JSON Schema format)
   • Input Form - Interactive form to test search parameters
   • Query - JSON representation of your search
   • Search Results - Records matching your criteria

​

Testing Process

1. Fill out the search form with various criteria
2. Click Search
3. Review the returned records
4. Adjust field configurations if results aren’t as expected
5. Click Copy & Test in Playground to test with OpenAI’s interface

​

Exporting Tables

• JSONL - Raw data, one record per line
• CSV - Spreadsheet-friendly format

• JSON - Complete table definition including field types and search configurations

​

Best Practices

​

Structure Your Data

• Use meaningful column names - Clear names help the AI understand the data
• Normalize values - Use consistent formats (e.g., all caps for status codes)
• Avoid empty fields - Fill in as many fields as possible for better filtering
• Include IDs - Add unique identifiers for precise record lookup
• Add timestamps - Include creation/update dates for temporal filtering

​

Choose Appropriate Search Types

• Text search - For fields users describe in various ways (product names, descriptions)
• Select/Multiselect - For fields with limited, known values (categories, statuses)
• Range - For continuous numbers (prices, quantities, scores)
• Between - For date ranges (order dates, event times)

​

Optimize for Performance

• Limit columns to relevant fields - More fields create more complex search forms
• Use select instead of text search when possible - Select fields are faster
• Keep data current - Update tables regularly by re-importing fresh data
• Test with realistic queries - Ensure common searches return results quickly

​

Write Clear Tool Descriptions

"Search our product catalog to find items by category, brand, price range,
and availability status. Use this when customers ask about products,
pricing, or inventory."

"Product search table"

• What data the table contains
• What fields can be searched
• When to use this table versus other tools

​

Common Use Cases

​

Product Catalog Search

• Import product data (SKU, name, category, brand, price, inventory)
• Configure text search on name and description
• Use select for category and brand
• Use range for price
• Use toggle for in-stock status

​

Customer Database Lookup

• Import customer data (ID, name, company, industry, location, status)
• Configure text search on name and company
• Use select for industry and location
• Use multiselect for status
• Include creation date with date filter

​

Order History Search

• Import order data (order ID, customer, product, quantity, price, date, status)
• Configure text search on order ID and customer
• Use select for status
• Use range for price and quantity
• Use date filter for order date

​

Knowledge Base Articles

• Import articles (title, content, category, tags, updated date)
• Configure text search on title and content
• Use multiselect for category and tags
• Use date filter for updated date

​

Updating Tables

1. Export your current table as JSON (to preserve field configuration)
2. Update the objects array in the JSON with new data
3. Import the updated JSON file
4. Verify data to ensure new records match field types

​

Troubleshooting

​

Table Not Appearing in AI Tools

• Verify the table has data loaded
• Check field configurations are complete
• Save any pending changes
• Refresh the AI profile page

​

AI Not Using the Table

• Review the tool description - make it more specific
• Test search manually to verify it works
• Check that search types match expected query patterns
• Add more descriptive field names

​

Validation Errors

• Review error messages to identify problematic rows
• Check data types match your data format
• Verify number formats match your settings
• Adjust field types to match actual data

​

Search Returns No Results

• Verify data is loaded (check row count)
• Test with broader search criteria
• Check field values match expected format
• Review filter combinations - they use AND logic

​

Performance Issues

• Reduce the number of searchable fields
• Use select fields instead of text search where possible
• Consider splitting large tables into smaller topic-specific tables
• Contact support for datasets over 500,000 rows

​

Advanced Features

​

Custom Field Names

• Original column name appears below
• Display name is shown in search forms
• Use descriptive names that clarify field purpose

​

Distribution Insights

• Click the lightbulb icon on any column
• See percentage and count for each value
• Helps identify data quality issues
• Assists in choosing search types

​

Pagination

• Use pagination controls at the bottom
• Select rows per page (10, 20, 50, 100)
• Jump to first/last page
• Navigate page by page

​

Cell Expansion

• Click any cell to expand it
• See full JSON objects
• View long text values
• Useful for complex data

​

Next Steps

• Instruct AI Agent - Configure AI profiles to use search tables
• Improve Answers - Optimize how the AI presents table results
• Data Providers - Combine tables with text-based knowledge sources