Count's AI Agent usage guide

This guide walks through how to add AI agents to your canvas and get the most from your collaboration with them.

Getting started guide

(1) Adding an agent to your canvas

Open any canvas where you want to analyze data
Click the Star icon in the toolbar
Select your data source from the prompt

Count AI will show suggested questions to get you started, or you can ask your own.

Tip: Use the microphone button to speak your question instead of typing—great for quickly sharing detailed context.

(2) Choosing data sources

When you add an agent, you'll select which data it can access. Valid sources include:

Catalogs (cached) - Recommended for most context-rich analysis
Database cells (cached)
DuckDB cells
CSV files in the canvas
Text objects like sticky notes (useful for adding context the agent should remember)

(3) Adding context to your questions

Make your questions more specific by referencing existing canvas objects:

Type @ and select catalog or cell names
Click the cursor button and select an individual cell
Click and drag to select multiple cells at once

The more context you provide, the better Count AI can answer your question.

(4) Working with agents in your canvas

Multiple agents: You can create multiple agents in one canvas to run different investigations simultaneously.

Collaborative: All analysts with the correct canvas permissions (data editors) can see every agent and continue conversations started by others, those without data editor permissions can see the conversations but not continue them or start new ones.

View progress: Click the blue dot on an agent to see:

Its thought process and progress
All questions asked by anyone in the canvas
How it's approaching your analysis

Move agents: Click and drag an agent to reposition it (and all its output) anywhere on the canvas.

(5) Understanding agent output

Count AI creates text and data objects in your canvas:

Low-code cells: When possible, the agent creates visual cells that clearly show where data came from—no SQL required.

SQL cells: For complex questions, the agent may write SQL. Click "Show input" on any cell to see the query it generated.

Visualizations: The agent can create charts and tables based on what makes sense for your question.

(6) Removing agents when you're done

You have two options:

Keep the agent: Leave it in place as a record of what you asked. You can return later to ask follow-up questions.

Delete the agent: Remove the agent dot while keeping all the cells and visualizations it created. Perfect when you've finished exploring and want to clean up your canvas.

Tips for better results

Be specific with your questions:

Instead of: "Show me sales data"
Try: "Show me monthly sales broken down by region for 2024"

Define custom metrics explicitly:

"Show me how margin (Profit / Sales) varies by customer"
"Calculate monthly retention as users active in month N who were also active in month N-1"

Use catalogs when possible: Catalogs contain rich metadata that helps Count AI understand your data model better than raw tables or CSVs.

Add persistent context: If there's information the agent needs to remember (definitions, business context, priorities), add it to a sticky note and include it when you create the agent. This saves you from repeating context in every question.

Ask follow-up questions: If the first answer isn't quite right, tell the agent what to adjust. It maintains context from your conversation.

What to do when Count AI can't help

Count AI works best with analytical questions about data you've added as context. It may struggle with:

Questions about data it doesn't have access to
Requests that require live database queries (it only works with cached data)
Very ambiguous questions without enough context

If you're not getting useful results, try:

Adding more specific context with @ mentions
Breaking complex questions into smaller steps
Defining any custom terms or metrics in your question