LLMs.txt directory

Supported data sources for agentic charts

Agentic charts are generated from the tabular output of analytical tools that produce non-token answers. This article describes each supported data source, what data it produces, and how it maps to charts.

Overview

Agentic charting supports any non-token answer that produces a tabular result set. The charting engine reads the result data, classifies columns by semantic type, and generates an appropriate visualization.

Source Description Typical use case

Code execution

Python-based code runs to analyze data

Forecasting, simulations, statistical modeling, custom calculations

MCP (Model Context Protocol)

Web search and external data retrieval

Comparing market data, retrieving public statistics, web-sourced analysis

AgentQL

Agent-generated data queries

Custom data analysis via agent-structured queries

Code execution

When you ask a question that requires computational analysis (for example, forecasting or trend analysis), Spotter can generate and execute Python code.

What data is produced

The code execution tool produces a tabular output with:

  • One or more data columns (quantitative values, dates, categories)

  • Optional metadata columns (confidence intervals, lower bounds, upper bounds)

  • Text output (explanations, interpretations)

The charting engine uses the tabular portion to render the chart.

Example

User asks:

"Forecast sales for the next 6 months"

Result:

Python code generates a table with: * Month (date) * Forecasted Sales (quantitative) * Lower Bound (quantitative) * Upper Bound (quantitative)

The charting engine renders a line chart with: * A line for the forecasted sales * Shaded confidence interval between the lower and upper bounds * Properly formatted time axis

When charts are not generated

If the code execution produces only text output (for example, a textual summary or explanation without tabular data), no chart is generated.

AgentQL

AgentQL is the agent’s structured query language for data retrieval.

What data is produced

AgentQL produces tabular results. The charting engine processes these results using the same classification and rendering pipeline.

Example

User asks:

"Analyze customer churn by region"

Result:

AgentQL retrieves a table with: * Region (text) * Churn Rate (numeric)

The charting engine renders a bar chart or map visualization depending on the data structure.

What makes data chartable

For agentic charting to generate a visualization, the non-token answer must produce:

  1. Tabular data — At least one row and one column of data

  2. At least one quantitative column — A numeric measure to plot

  3. At least one dimension column — A category, date, or label to organize the measure by

If the result contains only a single scalar value (for example, "Total revenue: $1,000,000"), the charting engine may display the value as a KPI or simple metric rather than a chart.

Related topics