Charts for agentic answers
Spotter now automatically visualizes non-token answers with interactive charts. Previously, answers generated through code execution, SpotQL, MCP, and other non-token paths returned data only in tabular or text format. With agentic charting, these answers are rendered as rich, interactive Muze charts directly within the Spotter conversation.
What is a non-token answer?
In Spotter, answers are produced through different tools:
- Token-based answers (TML answers)
-
Generated from ThoughtSpot’s native data models, worksheets, and tables. These answers already support ThoughtSpot’s standard charting.
- Non-token answers
-
Generated from analytical tools that run outside of ThoughtSpot’s native data model. These include:
-
Code execution (Python-based analysis)
-
SpotQL (SQL-based data queries)
-
MCP (Model Context Protocol) tool output
-
AgentQL and other agent-generated data
-
Non-token answers previously returned data as plain tables or text. Agentic charting closes this gap by automatically rendering charts from the tabular output of these tools.
How agentic charting works
When you ask a data question that triggers a non-token answer, the following happens automatically:
-
Data generation — The analytical tool (for example, code execution or SpotQL) produces tabular data and persists it.
-
Schema classification — An LLM-powered schema pipeline classifies each column by type (quantitative, categorical, date, ordinal, geographic) and determines appropriate aggregations, time buckets, and formatting.
-
Chart generation — The charting tool calls the Muze charting service to generate an interactive chart based on the semantic meaning of your question and the data structure.
-
Rendering — The chart is embedded in the agent’s response and displayed in the conversation.
The chart is automatically generated on the same turn as the answer, before the final response is delivered.
Key features
Intent-aware chart generation
The chart is composed from the semantic meaning of your question and the result data, not just raw column types. For example, when you ask for a forecast for the next 6 months, the charting engine understands this is a time-series forecast and renders an appropriate chart with forecasted values and confidence intervals.
AI-powered chart recommendations and enrichment
Charts are automatically enriched with:
-
Recommended chart types based on data structure
-
Appropriate encoding, formatting, and styling
-
Configured axes, legends, and color schemes
-
Data labels and annotations where relevant
Natural-language chart editing
After a chart is generated, you can refine it using natural-language follow-up commands. The system interprets your request and re-renders the chart accordingly. See Edit charts with natural language for a full list of supported commands and examples.
Progressive charting
The charting system supports a progressive rendering experience. A base chart is rendered within 3–5 seconds, and enhancements (styling, annotations, formatting) are applied in the background and merged seamlessly once complete. See Progressive charting for details.
When to use agentic charts
Use agentic charting when you want to:
-
Visualize data produced by Python code execution (for example, statistical forecasts, simulations, or custom calculations)
-
Chart results from external web data retrieved through MCP
-
Display SpotQL query results visually
-
Create charts from AgentQL or other agent-generated data
Agentic charts are not used for standard token-based answers from ThoughtSpot worksheets and tables. Those continue to use ThoughtSpot’s existing charting system.