Components as Tools

Let your agent render rich React components directly in the chat by calling them as tools.

"""PydanticAI agent backing the Tool-Based Generative UI demo.Mirrors showcase/integrations/langgraph-python/src/agents/gen_ui_tool_based.py.The frontend registers `render_bar_chart` and `render_pie_chart` tools via`useComponent`. CopilotKit's runtime injects those tool definitions into theagent request at runtime, so the agent does not need to declare them locally —PydanticAI's AG-UI bridge surfaces frontend-registered tools to the model oneach run, and the model decides when to call them."""from __future__ import annotationsfrom textwrap import dedentfrom pydantic_ai import Agentfrom pydantic_ai.models.openai import OpenAIResponsesModelSYSTEM_PROMPT = dedent(    """    You are a data visualization assistant.    When the user asks for a chart, call `render_bar_chart` or    `render_pie_chart` with a concise title, short description, and a    `data` array of `{label, value}` items. Pick bar for comparisons over    a small set of categories; pick pie for composition / share-of-whole.    Keep chat responses brief — let the chart do the talking.    """).strip()agent = Agent(    model=OpenAIResponsesModel("gpt-4.1-mini"),    system_prompt=SYSTEM_PROMPT,)

What is this?#

Tool-based Generative UI is the simplest form of Generative UI: you register a React component with useComponent, and CopilotKit exposes it to the agent as a tool. When the agent calls the tool, CopilotKit renders your component inline in the chat, passing the tool's arguments straight through as typed props.

Unlike tool rendering, which wraps a real backend tool in a custom UI, tool-based GenUI is the component. There is no handler, no user interaction, no server-side execution. The agent decides when to show it, populates the data, and CopilotKit paints it.

When should I use this?#

Use useComponent when you want to:

Display rich UI (cards, charts, tables, dashboards) inline in the chat
Show structured data the agent has derived from its reasoning
Render previews, status indicators, or visual summaries
Let the agent present information beyond plain text

For components that need user interaction, see Human-in-the-loop. For operational transparency around a real backend tool, see Tool rendering.

How it works in code#

Nothing to wire on the agent

PydanticAI's AG-UI bridge surfaces frontend-registered tools to the model on every run, so the agent declares no tools of its own. A component registered with useComponent reaches the model through the AG-UI request payload and the model calls it by name.

src/agents/chart_agent.py

from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIResponsesModel

agent = Agent(
    model=OpenAIResponsesModel("gpt-4.1-mini"),
    system_prompt=SYSTEM_PROMPT,
)

Tell the model when to call it

This is the part that is easy to miss. The tool arrives on every run, but a model with no instruction about it will answer in prose and never call it. Name the tool in the system prompt and say what it is for.

src/agents/chart_agent.py

SYSTEM_PROMPT = """
You are a data visualization assistant.

When the user asks for a chart, call `render_bar_chart` with a concise
title and a `data` array of `{label, value}` items.

Keep chat responses brief — let the chart do the talking.
"""

useComponent takes a name, a Zod schema for its props, and the component to render. The runtime registers it as a frontend tool so the agent can discover it, and Zod validates the LLM's arguments before they reach your component.

page.tsx

  useComponent({    name: "render_bar_chart",    description: "Display a bar chart with labeled numeric values.",    parameters: barChartPropsSchema,    render: BarChart,  });

The component itself is ordinary React: it reads only its props and can stream in as the agent fills the payload. The example above uses Recharts for the bar chart; it doesn't know anything about CopilotKit.

The name you pass to useComponent is what the agent sees as the tool name. Make it a verb like render_bar_chart or show_weather so the LLM reliably picks it when the user asks for that visualization.