Reasoning Messages

Customize how reasoning (thinking) tokens from models like o1, o3, and o4-mini are displayed.

"""Reasoning-capable PydanticAI agent for the reasoning family of demos.Backs two showcase cells:    - agentic-chat-reasoning       (custom amber ReasoningBlock slot)    - reasoning-default-render     (CopilotKit's built-in reasoning card)Mirrors `showcase/integrations/langgraph-python/src/agents/reasoning_agent.py`(shared across reasoning demos there).Why a reasoning model:PydanticAI's AG-UI bridge surfaces reasoning summaries from``OpenAIResponsesModel`` as ``THINKING_*`` / ``REASONING_*`` events on theAG-UI stream when the underlying OpenAI Responses API returns reasoningitems. The Responses API only returns reasoning content for nativereasoning models (gpt-5, o3, o4-mini, etc.) — gpt-4o / gpt-4.1 do notemit reasoning items, so a non-reasoning model would never produce theevents the frontend renders. We therefore pin a reasoning model here.Model choice:``gpt-5`` is the team default for reasoning-bearing showcase cells. It isthe latest OpenAI reasoning-capable model available through the ResponsesAPI and emits reasoning summaries that the AG-UI bridge translates toTHINKING events."""from __future__ import annotationsimport osfrom textwrap import dedentfrom pydantic_ai import Agentfrom pydantic_ai.models.openai import OpenAIResponsesModelfrom pydantic_ai.models.openai import OpenAIResponsesModelSettingsSYSTEM_PROMPT = dedent(    """    You are a helpful assistant. For each user question, first think    step-by-step about the approach, then give a concise answer.    Keep thinking concise — two to four short steps is plenty for most    questions. Do not repeat the final answer inside the reasoning block.    """).strip()# Auto-summary so the Responses API surfaces reasoning content as# reasoning items (which the AG-UI bridge then forwards as THINKING /# REASONING events). Without a summary the API can omit the reasoning# payload from the streamed response._REASONING_MODEL = os.environ.get("REASONING_MODEL", "gpt-5")agent = Agent(    model=OpenAIResponsesModel(_REASONING_MODEL),    model_settings=OpenAIResponsesModelSettings(        openai_reasoning_summary="auto",    ),    system_prompt=SYSTEM_PROMPT,)

Some models (like OpenAI's o1, o3, and o4-mini) emit reasoning tokens: internal "thinking" traces that show the model's chain-of-thought before it produces a final answer. CopilotKit surfaces these tokens automatically with a collapsible Reasoning Message card.

Default Behavior#

When reasoning events arrive from the agent, CopilotKit renders them inside a built-in card that:

Shows a "Thinking…" label with a pulsating indicator while the model is reasoning.
Expands automatically so you can follow the model's thought process in real-time.
Collapses and switches to "Thought for X seconds" once reasoning finishes.
Renders the reasoning content as Markdown.
Includes a chevron toggle so users can re-expand and review the reasoning at any time.

No extra configuration is needed; if your model emits reasoning tokens, the card appears automatically.

The only requirement is connecting your agent to CopilotKit; no extra props or configuration needed:

page.tsx

          <CopilotChat            agentId="reasoning-default-render"            className="h-full rounded-2xl"          />

Customizing the Reasoning Message#

The reasoning message is composed of three sub-components that can each be replaced independently via slot props:

Sub-component	Slot prop	Description
`Header`	`header`	The clickable bar with the brain icon, label, and chevron
`Content`	`contentView`	The reasoning text area (Markdown)
`Toggle`	`toggle`	The expand/collapse animation wrapper

You pass custom sub-components through the messageView prop on CopilotChat, CopilotPopup, or CopilotSidebar:

<CopilotChat
  messageView={{
    reasoningMessage: {
      header: CustomHeader,
      contentView: CustomContent,
    },
  }}
/>

Custom Header#

Replace the header to change the icon, label text, or styling. The header receives these props:

Prop	Type	Description
`isOpen`	`boolean`	Whether the content panel is currently expanded
`label`	`string`	`"Thinking…"` while streaming, `"Thought for X seconds"` after
`hasContent`	`boolean`	Whether any reasoning text has been received
`isStreaming`	`boolean`	Whether reasoning is actively streaming
`onClick`	`() => void`	Toggle handler (only present when `hasContent` is `true`)

import { CopilotChat } from "@copilotkit/react-core/v2";
import "@copilotkit/react-core/v2/styles.css";

function CustomHeader({
  isOpen,
  label,
  hasContent,
  isStreaming,
  ...props
}: React.ButtonHTMLAttributes<HTMLButtonElement> & {
  isOpen?: boolean;
  label?: string;
  hasContent?: boolean;
  isStreaming?: boolean;
}) {
  return (
    <button
      className="flex w-full items-center gap-2 px-3 py-2 text-sm font-medium"
      {...props}
    >
      {isStreaming ? "🧠" : "💡"}
      <span>{label}</span>
      {hasContent && (
        <span className="ml-auto text-xs">{isOpen ? "Hide" : "Show"}</span>
      )}
    </button>
  );
}

<CopilotChat
  messageView={{
    reasoningMessage: { header: CustomHeader },
  }}
/>

Custom Content#

Replace the content area to change how reasoning text is displayed:

Prop	Type	Description
`isStreaming`	`boolean`	Whether reasoning tokens are still arriving
`hasContent`	`boolean`	Whether any reasoning text has been received
`children`	`string`	The raw reasoning text

function CustomContent({
  isStreaming,
  hasContent,
  children,
  ...props
}: React.HTMLAttributes<HTMLDivElement> & {
  isStreaming?: boolean;
  hasContent?: boolean;
}) {
  if (!hasContent && !isStreaming) return null;

  return (
    <div className="px-4 pb-3 text-sm text-gray-500 font-mono" {...props}>
      {children}
      {isStreaming && <span className="animate-pulse ml-1">▊</span>}
    </div>
  );
}

<CopilotChat
  messageView={{
    reasoningMessage: { contentView: CustomContent },
  }}
/>

Fully Custom Reasoning Message#

For complete control over the entire reasoning card, pass a component instead of slot props. Your component receives the same top-level props as the built-in one:

Prop	Type	Description
`message`	`ReasoningMessage`	The reasoning message object (`.content` holds the text)
`messages`	`Message[]`	All messages in the conversation
`isRunning`	`boolean`	Whether the agent is currently running

"""Reasoning-capable PydanticAI agent for the reasoning family of demos.Backs two showcase cells:    - agentic-chat-reasoning       (custom amber ReasoningBlock slot)    - reasoning-default-render     (CopilotKit's built-in reasoning card)Mirrors `showcase/integrations/langgraph-python/src/agents/reasoning_agent.py`(shared across reasoning demos there).Why a reasoning model:PydanticAI's AG-UI bridge surfaces reasoning summaries from``OpenAIResponsesModel`` as ``THINKING_*`` / ``REASONING_*`` events on theAG-UI stream when the underlying OpenAI Responses API returns reasoningitems. The Responses API only returns reasoning content for nativereasoning models (gpt-5, o3, o4-mini, etc.) — gpt-4o / gpt-4.1 do notemit reasoning items, so a non-reasoning model would never produce theevents the frontend renders. We therefore pin a reasoning model here.Model choice:``gpt-5`` is the team default for reasoning-bearing showcase cells. It isthe latest OpenAI reasoning-capable model available through the ResponsesAPI and emits reasoning summaries that the AG-UI bridge translates toTHINKING events."""from __future__ import annotationsimport osfrom textwrap import dedentfrom pydantic_ai import Agentfrom pydantic_ai.models.openai import OpenAIResponsesModelfrom pydantic_ai.models.openai import OpenAIResponsesModelSettingsSYSTEM_PROMPT = dedent(    """    You are a helpful assistant. For each user question, first think    step-by-step about the approach, then give a concise answer.    Keep thinking concise — two to four short steps is plenty for most    questions. Do not repeat the final answer inside the reasoning block.    """).strip()# Auto-summary so the Responses API surfaces reasoning content as# reasoning items (which the AG-UI bridge then forwards as THINKING /# REASONING events). Without a summary the API can omit the reasoning# payload from the streamed response._REASONING_MODEL = os.environ.get("REASONING_MODEL", "gpt-5")agent = Agent(    model=OpenAIResponsesModel(_REASONING_MODEL),    model_settings=OpenAIResponsesModelSettings(        openai_reasoning_summary="auto",    ),    system_prompt=SYSTEM_PROMPT,)

The ReasoningBlock used above renders the reasoning as an amber-tagged inline banner, intentionally louder than the default card so the thinking chain is the focal UI of the demo. Swap in your own component to match your product's tone:

page.tsx

import React from "react";import {  CopilotKit,  CopilotChat,  CopilotChatReasoningMessage,} from "@copilotkit/react-core/v2";import { ReasoningBlock } from "./reasoning-block";export default function AgenticChatReasoningDemo() {  return (    <CopilotKit runtimeUrl="/api/copilotkit" agent="agentic-chat-reasoning">      <div className="flex justify-center items-center h-screen w-full">        <div className="h-full w-full max-w-4xl">          <Chat />        </div>      </div>    </CopilotKit>  );}function Chat() {  return (    <CopilotChat      agentId="agentic-chat-reasoning"      className="h-full rounded-2xl"      messageView={{        reasoningMessage: ReasoningBlock as typeof CopilotChatReasoningMessage,      }}    />  );}

Render-Prop Children#

The built-in CopilotChatReasoningMessage also supports a render-prop pattern for cases where you want to rearrange the built-in sub-components without reimplementing them:

import {
  CopilotChatReasoningMessage,
} from "@copilotkit/react-core/v2";
import { CopilotChat } from "@copilotkit/react-core/v2";
import "@copilotkit/react-core/v2/styles.css";

function MyReasoningLayout(props: React.ComponentProps<typeof CopilotChatReasoningMessage>) {
  return (
    <CopilotChatReasoningMessage {...props}>
      {({ header, toggle }) => (
        <div className="rounded-lg border bg-yellow-50 my-2">
          {header}
          {toggle}
        </div>
      )}
    </CopilotChatReasoningMessage>
  );
}

<CopilotChat
  messageView={{
    reasoningMessage: MyReasoningLayout,
  }}
/>

The render-prop callback receives:

Property	Description
`header`	Pre-rendered header element
`contentView`	Pre-rendered content element
`toggle`	Pre-rendered expand/collapse wrapper (contains `contentView`)
`message`	The reasoning message object
`messages`	All messages
`isRunning`	Whether the agent is running