MetricChat
Using MetricChat

Data Sources

MetricChat connects to virtually any data infrastructure — databases, warehouses, and business applications — for AI-powered analytics.

The system queries your data in place, preserving existing security controls while enabling AI agents to understand schema and relationships.

Supported Connections

Databases

  • PostgreSQL
  • MySQL
  • MariaDB
  • Microsoft SQL Server
  • Oracle DB
  • ClickHouse
  • MongoDB

Cloud Warehouses

  • Snowflake
  • Google BigQuery
  • AWS Redshift
  • Azure Data Explorer
  • Apache Pinot

Business Applications

  • Salesforce
  • NetSuite
  • Tableau

File-Based Sources

  • CSV / Excel upload — Upload spreadsheets to create instant queryable DuckDB data sources
  • PDF upload — Upload documents for AI-assisted analysis
  • DuckDB

Connecting a Data Source

  1. Navigate to Settings > Data Sources
  2. Click Add Connection
  3. Select the data source type
  4. Enter connection credentials
  5. Test the connection
  6. Select which tables the AI can access

Uploading Files

For CSV and Excel files:

  1. Click Create Database in the connection flow
  2. Upload one or more files
  3. MetricChat automatically creates a DuckDB data source with your data
  4. Tables are immediately available for AI queries

Configuration

Table Management

Selectively enable or disable table visibility to AI agents. Start with high-priority tables and expand access as needed.

Instructions and Context

Enrich the AI's understanding by:

  • Importing documentation from Git repositories (dbt, Tableau, LookML, markdown)
  • Adding source-specific instructions for business rules
  • Defining relationships between tables

Access Control

Control which team members can access each data source. Optional user authentication requirements maintain data security while enabling collaboration.

Best Practices

Write Good Descriptions

Detailed data source descriptions explaining business domains, key entities, and data structure significantly improve AI performance.

Example:

"This database contains our movie rental business data. Key tables include film (catalog), rental (transactions), customer (profiles), and payment (billing). Films connect to categories and actors through junction tables."

Conversation Starters

Add pre-defined prompts that appear as interactive chips, helping users quickly explore common analyses relevant to their data.

Recommendations

  • Provide comprehensive context about data domains and relationships
  • Enable only relevant tables to avoid overwhelming the AI
  • Create descriptive connection names
  • Implement access controls for sensitive data
  • Develop source-specific instructions for business rules

On this page