A widget agent is an AI Agent component that creates individual data visualizations from natural language descriptions. Tell it “show me monthly revenue vs. customer count as a combo chart” and it selects the chart type, maps revenue to the primary axis, customer count to the secondary axis, detects the date field for the time dimension, and renders the visualization. Widget agents handle the technical translation between what you want to see and how the BI platform needs to be configured to show it.
TL;DR
- A widget agent creates individual charts and visualizations from plain English descriptions.
- Core capability: intelligent field mapping, automatically assigning dataset columns to the right chart properties (axes, dimensions, measures, colors).
- Detects date/time fields automatically for time-series configurations.
- Supports combo charts with secondary axes, multi-series visualizations, and conditional formatting.
- Works within the agent orchestration system: the Query Agent finds the data, the Widget Agent visualizes it, the Dashboard Agent arranges the layout.
- Available in-platform, via MCP, and as an embeddable component for SaaS products.
Table fo Contents
Why Widget Agents Exist
Creating a chart in a traditional BI tool requires knowing which fields go where. You need to understand the difference between a dimension and a measure, know that dates belong on the X-axis, choose between a bar and a line chart, and configure aggregation functions. For analysts, this is routine. For everyone else, it is a barrier.
Widget agents remove this barrier. They understand data semantics well enough to make these decisions automatically. A field called “revenue” gets treated as a measure. A field called “created_date” gets detected as a time dimension. A dataset with geographic fields triggers a map visualization.
How Widget Agents Work
Intelligent Field Mapping
When a widget agent receives a request, it reads the dataset schema and maps fields to chart properties using contextual understanding:
- Measures: Numeric fields like revenue, count, percentage are mapped to Y-axes or value properties
- Dimensions: Categorical fields like product name, region, status are mapped to X-axes or group-by properties
- Temporal: Date and timestamp fields are automatically detected and placed on time axes with appropriate granularity (daily, weekly, monthly)
- Geographic: Fields containing country, state, city, or coordinate data trigger map visualizations
Chart Type Selection
The agent selects chart types based on the data and the user’s intent. “Compare revenue by region” produces a bar chart. “Show revenue over time” produces a line chart. “Revenue by region over time” produces a multi-series line or area chart. Users can override with explicit requests: “make it a heatmap” or “show as a treemap.”
Multi-Chart and Combo Support
Widget agents handle complex visualizations that typically require advanced BI knowledge. A combo chart with revenue on the primary axis and growth rate on the secondary axis, using bars for revenue and a line for growth rate, is a single natural language request: “show revenue as bars and growth rate as a line on the same chart.”
Widget Agent vs. Dashboard Agent
These two agents serve different purposes and work together:
| Aspect | Widget Agent | Dashboard Agent |
| Scope | Creates and configures a single visualization | Creates and manages entire dashboards (multiple widgets + layout) |
| Input | Dataset ID + description of desired chart | Natural language description of an analytical view |
| Field mapping | Maps dataset columns to chart axes, colors, sizes | Delegates field mapping to Widget Agent for each chart |
| Layout | Not applicable (single widget) | Auto-layout of multiple widgets in a grid |
| Modification | Change chart type, colors, labels, axes | Add filters, generate share URLs, export to PDF/Excel/PPT |
| When invoked | When user asks for a specific chart or visualization | When user asks for a dashboard, a view, or a collection of charts |
| Orchestration | Often invoked BY the Dashboard Agent as a sub-step | Coordinates with Widget Agent and Recommendation Agent |
In practice, asking “create a sales dashboard” invokes the Dashboard Agent, which then calls the Widget Agent multiple times to create each individual chart. Asking “add a bar chart of revenue by product to this dashboard” invokes the Widget Agent directly.
Widget Agents in Traditional BI vs. Agentic BI
| Task | Traditional BI (manual) | Agentic BI (widget agent) |
| Create a bar chart | Select dataset, drag dimension to X-axis, drag measure to Y-axis, choose chart type, configure aggregation | “Show revenue by region as a bar chart” |
| Create a combo chart | Select dual-axis option, assign primary/secondary measures, choose bar vs. line per series, adjust scales | “Revenue as bars, margin as a line, same chart” |
| Change chart type | Click chart type menu, select new type, re-verify field mappings | “Change this to a pie chart” |
| Add conditional formatting | Open formatting panel, define rules, set thresholds, choose colors | “Highlight cells red where churn exceeds 5%” |
| Time needed | 2-10 minutes per widget depending on complexity | 5-15 seconds |
Embedding Widget Agents
For SaaS companies embedding analytics in their product, the widget agent is available as part of Knowi’s embedded analytics platform. End users in your application can create their own visualizations through natural language without your team building a chart builder UI.
Multi-tenant isolation ensures each customer only visualizes their own data. Row-level security, white-label branding, and role-based permissions all apply. The widget agent can be restricted to specific chart types or datasets per customer tier.
Other Agents with AgenticBI
Read more about other agents that we are building under Agentic BI:
Frequently Asked Questions
What is a widget agent?
A widget agent is an AI component that creates individual data visualizations (charts, tables, maps) from natural language descriptions. It handles field mapping, chart type selection, and configuration automatically, so users describe what they want to see instead of manually building charts in a BI tool.
How does a widget agent decide which chart type to use?
The agent analyzes the data types in your request and dataset. Comparisons across categories produce bar charts. Trends over time produce line charts. Part-to-whole relationships produce pie or treemap charts. Geographic data produces maps. Users can override any automatic selection with explicit instructions.
What is intelligent field mapping?
Intelligent field mapping is the agent’s ability to automatically assign dataset columns to the correct chart properties. It detects which fields are measures (numeric values like revenue), dimensions (categories like region), and temporal (dates). This eliminates the need to manually drag fields to axes in a chart builder.
Can a widget agent create combo charts with dual axes?
Yes. Describe what you want on each axis: “show revenue as bars on the left axis and growth rate as a line on the right axis.” The agent configures the primary and secondary axes, assigns the correct chart type to each series, and scales them independently.
How does the widget agent work with the dashboard agent?
The dashboard agent manages entire dashboards (layout, filtering, sharing). When it needs to create individual charts, it calls the widget agent for each one. The widget agent creates and configures the visualization; the dashboard agent handles placement and layout. They communicate through a shared orchestration context.
Where can I try agents for my business?
Try AI agents for your data at AgenticBI.com starting at $99/month, or schedule a demo for enterprise usecase.