Skip to main content

AI Agents: Overview

Datalinx includes three specialized AI agents that assist you throughout the platform. They live in the chat panel on the left side of your screen and switch automatically based on which section of the platform you're using.

The Three Agents​

Analysis Agent​

Where it appears: Dashboard, Explore (Explore, Insights, Activate)

What it does: Helps you understand and analyze your data. The Analysis Agent is your go-to for:

  • Writing SQL queries against your data
  • Generating visualizations and charts
  • Creating AI-powered insights
  • Building audience segments
  • Answering ad-hoc business questions

Example conversations:

  • "Show me the top 10 customers by total spend"
  • "What's the trend in new signups over the past 6 months?"
  • "Create an insight about customer churn drivers"
  • "Build an audience of users who haven't logged in for 30 days"

Pipeline Agent​

Where it appears: Configure (Data Sources, Design, Transform, Test, Monitor, Connect)

What it does: Helps you build and manage your data pipeline. The Pipeline Agent handles the technical work of:

  • Discovering and scanning source tables
  • Designing target schemas
  • Creating field mappings and CTEs
  • Configuring data quality tests
  • Troubleshooting pipeline issues
  • Managing GitHub integration

Example conversations:

  • "Discover the tables in my source database"
  • "Create a target schema for customer analytics"
  • "Map the users table to the customers target"
  • "Why is the email mapping showing an error?"
  • "Run the tests for the orders table"

Foundations Agent​

Where it appears: Foundations (Object Types, Data Types, Metrics, Business Logic, Identity)

What it does: Helps you build and maintain your semantic layer. The Foundations Agent specializes in:

  • Scanning databases to auto-generate entity definitions
  • Creating and editing Object Types and their relationships
  • Defining business metrics and KPIs
  • Configuring identity resolution rules
  • Setting up business logic rules

Example conversations:

  • "Scan my source database and identify all business entities"
  • "Create a Customers object type with standard attributes"
  • "Define a Total Revenue metric as the sum of completed order amounts"
  • "Set up identity resolution for email and device ID"

How the Chat Panel Works​

Sending Messages​

Type your question or instruction in the text box at the bottom of the chat panel and press Enter (or click Send). The agent will process your request and respond.

Streaming Responses​

Responses stream in real-time — you'll see the text appear as the agent generates it. For longer responses, this means you can start reading before the full answer is complete.

Tool Calls​

When an agent takes an action (like creating a mapping or running a query), you'll see a tool call in the chat. Tool calls show:

  • Tool name — what action is being performed
  • Status icon — running (spinner), succeeded (checkmark), or failed (X)
  • Input — what parameters were passed to the tool
  • Result — what the tool returned

You can expand tool calls to see the full details, and copy results to your clipboard.

Approval Requests​

For actions that modify your data or configuration, the agent will ask for your approval before proceeding. You'll see an approval panel with:

  • A description of what the agent wants to do
  • Approve and Deny buttons

This ensures the agent never makes changes you haven't authorized.

Conversation History​

Your chat history is preserved within your session. You can scroll up to review previous messages, tool calls, and results. The agent maintains context throughout the conversation, so you can refer back to earlier topics.

Cancelling Requests​

If the agent is taking too long or going in the wrong direction, you can cancel the current request using the stop button that appears while the agent is processing.

Tips for Working with Agents​

  • Be specific — "Create a customers table with email, name, and signup_date columns" works better than "Set up my data"
  • Iterate — Start with a request, review what the agent produces, and refine with follow-up messages
  • Review tool calls — When the agent takes actions, expand the tool call to verify what was done
  • Use approval wisely — The approval system is there to protect you. Take a moment to review what the agent wants to do before approving.
  • Switch context — If you need a different agent, just navigate to the appropriate section of the platform. The agent panel will switch automatically.
  • Ask for explanations — The agents can explain why they're recommending a particular approach. Don't hesitate to ask "Why?" or "Can you explain that?"

Workspace Prompt​

Each workspace can have a custom workspace prompt — additional instructions that are given to the Pipeline Agent for every conversation. This is useful for:

  • Providing context about your specific data model
  • Setting naming conventions
  • Specifying business rules the agent should follow

You can edit the workspace prompt in your workspace settings.

Last updated on