ContextQMD
Back to Copilotkit

useCopilotKit

showcase/shell-docs/src/content/reference/hooks/useCopilotKit.mdx

1.57.06.4 KB
Original Source

Overview

useCopilotKit is a low-level React hook that returns the CopilotKit context value, providing direct access to the core instance and provider-level state. It subscribes to runtime connection status changes and triggers re-renders when the connection status updates.

<Callout type="info"> `useCopilotKit` is a low-level hook. Most applications should use higher-level hooks like `useAgent` or `useFrontendTool` instead. </Callout>

Throws an error if used outside of a CopilotKitProvider (or the <CopilotKit> wrapper component).

Signature

tsx
import { useCopilotKit } from "@copilotkit/react-core/v2";

function useCopilotKit(): CopilotKitContextValue;

Parameters

This hook takes no parameters.

Return Value

<PropertyReference name="context" type="CopilotKitContextValue"> The context object containing the core instance and provider-level state. <PropertyReference name="copilotkit" type="CopilotKitCoreReact"> The live CopilotKit core instance. This is the central coordinator that manages agents, tools, suggestions, and runtime communication. It exposes methods for: - Agent management (`agents`, `getAgent`) - Tool registration and execution - Suggestion configuration and retrieval - Runtime connection management - Event subscription via `subscribe()` 
The core instance is created once per `CopilotKitProvider` and is stable across re-renders.
</PropertyReference> <PropertyReference name="executingToolCallIds" type="ReadonlySet<string>"> Set of tool call IDs currently being executed. This is tracked at the provider level to ensure tool execution events are captured even before child components mount. This is important for scenarios like human-in-the-loop reconnection, where pending tool calls may already exist when the component tree mounts. </PropertyReference> </PropertyReference>

Usage

Accessing the Core Instance

tsx
function DebugPanel() {
  const { copilotkit } = useCopilotKit();

  const agents = Object.keys(copilotkit.agents ?? {});

  return (
    <div>
      <h3>Registered Agents</h3>
      <ul>
        {agents.map((id) => (
          <li key={id}>{id}</li>
        ))}
      </ul>
    </div>
  );
}

Subscribing to Core Events

tsx
function ConnectionMonitor() {
  const { copilotkit } = useCopilotKit();

  useEffect(() => {
    const subscription = copilotkit.subscribe({
      onRuntimeConnectionStatusChanged: () => {
        console.log("Runtime connection status changed");
      },
    });

    return () => {
      subscription.unsubscribe();
    };
  }, [copilotkit]);

  return null;
}

Running a Tool Programmatically

copilotkit.runTool() lets you execute a registered frontend tool directly from code — no LLM turn required. The tool's handler runs, render components appear in the UI, and both the tool call and result are added to the agent's message history.

tsx
function ExportButton() {
  const { copilotkit } = useCopilotKit();

  // Register the tool
  useFrontendTool({
    name: "exportData",
    description: "Export data as CSV",
    parameters: [{ name: "format", type: "string" }],
    handler: async ({ format }) => {
      const csv = await generateCsv(format);
      downloadFile(csv);
      return `Exported as ${format}`;
    },
  });

  // Trigger it from a button — no LLM needed
  const handleExport = async () => {
    const { result, error } = await copilotkit.runTool({
      name: "exportData",
      parameters: { format: "csv" },
    });
    if (error) console.error(error);
  };

  return <button onClick={handleExport}>Export CSV</button>;
}

runTool Parameters

<PropertyReference name="params" type="CopilotKitCoreRunToolParams"> <PropertyReference name="name" type="string" required> Name of the registered frontend tool to execute. </PropertyReference> <PropertyReference name="agentId" type="string"> Agent ID for tool and message scoping. Defaults to `"default"`. </PropertyReference>

<PropertyReference name="parameters" type="Record<string, unknown>" default="{}"

Arguments passed to the tool handler. </PropertyReference>

<PropertyReference name="followUp" type="string | false" default="false"> Controls whether an LLM follow-up is triggered after execution: - `false` — execute tool and stop (default) - `"generate"` — trigger an agent run so the LLM responds to the tool result - Any other string — add a user message with this text, then trigger an agent run </PropertyReference> </PropertyReference>

runTool Return Value

<PropertyReference name="result" type="CopilotKitCoreRunToolResult"> <PropertyReference name="toolCallId" type="string"> Unique ID of the tool call, matching the message added to history. </PropertyReference> <PropertyReference name="result" type="string"> Stringified return value from the tool handler (empty string for render-only tools). </PropertyReference> <PropertyReference name="error" type="string | undefined"> Error message if the handler threw. The tool result message will contain `"Error: ..."`. </PropertyReference> </PropertyReference>

Checking Tool Execution State

tsx
function ToolExecutionIndicator() {
  const { executingToolCallIds } = useCopilotKit();

  if (executingToolCallIds.size === 0) {
    return null;
  }

  return <div>Executing {executingToolCallIds.size} tool call(s)...</div>;
}

Behavior

  • Error on Missing Provider: Throws an error if the hook is used outside of CopilotKitProvider or the <CopilotKit> wrapper component.
  • Runtime Status Subscription: The hook subscribes to onRuntimeConnectionStatusChanged events, so components re-render when the runtime connection completes or fails.
  • Stable Core Reference: The copilotkit instance is created once per provider and remains stable across re-renders. Only the executingToolCallIds set changes as tool calls begin and complete.
  • Provider-Level Tool Tracking: executingToolCallIds is tracked at the provider level rather than in individual components. This ensures that tool execution start events fired before child components mount are not lost.