copilotkit-langgraph-history

Open-source LangGraph thread history persistence for CopilotKit. Restore chat history on page refresh with your own LangGraph deployment.

The Problem

Building a chat app with CopilotKit and LangGraph? You've probably noticed:

• Page refresh = empty chat - All messages disappear
• Thread switching loses context - No automatic history restoration
• No persistence of agent state - Users lose context

CopilotKit's default runtime doesn't automatically fetch historical messages from LangGraph's checkpoint system. This package adds that capability.

The Solution

With this package:

• Chat history restored on page load
• Seamless thread switching
• Agent state preserved
• Works with any LangGraph deployment (LangGraph Cloud, self-hosted)
• MIT licensed, open-source

Installation

npm install copilotkit-langgraph-history
# or
pnpm add copilotkit-langgraph-history
# or
yarn add copilotkit-langgraph-history

Peer Dependencies

This package requires the following peer dependencies:

npm install @copilotkit/runtime @copilotkitnext/runtime @ag-ui/core @langchain/langgraph-sdk rxjs

Quick Start

Next.js App Router

// app/api/copilotkit/route.ts
import { CopilotRuntime, createCopilotEndpointSingleRoute } from "@copilotkit/runtime/v2";
import {
  HistoryHydratingAgentRunner,
  createIsolatedAgent,
} from "copilotkit-langgraph-history";

const deploymentUrl = process.env.LANGGRAPH_DEPLOYMENT_URL!;
const langsmithApiKey = process.env.LANGSMITH_API_KEY;
const graphId = "my-agent";

function createRuntime() {
  // Create isolated agent (prevents serverless state contamination)
  const agent = createIsolatedAgent({
    deploymentUrl,
    graphId,
    langsmithApiKey,
  });

  // Create history-hydrating runner
  const runner = new HistoryHydratingAgentRunner({
    agent,
    deploymentUrl,
    graphId,
    langsmithApiKey,
    historyLimit: 100, // Max messages to load
  });

  return new CopilotRuntime({
    agents: { [graphId]: agent },
    runner,
  });
}

export const POST = async (req: Request) => {
  const runtime = createRuntime();
  const route = createCopilotEndpointSingleRoute({
    runtime,
    basePath: "/api/copilotkit",
  });
  return route.handleRequest(req);
};

Frontend (React)

import { CopilotKit } from "@copilotkit/react-core";
import { CopilotChat } from "@copilotkit/react-ui";

function App() {
  return (
    <CopilotKit
      runtimeUrl="/api/copilotkit"
      agent="my-agent"
      threadId={threadId} // Pass your thread ID here
    >
      <CopilotChat />
    </CopilotKit>
  );
}

Configuration

HistoryHydratingAgentRunner Options

Option	Type	Default	Description
agent	LangGraphAgent	required	The LangGraphAgent instance
deploymentUrl	string	required	LangGraph deployment URL
graphId	string	required	Graph identifier
langsmithApiKey	string	undefined	LangSmith API key
historyLimit	number	100	Max checkpoints to fetch (max 1000)
clientTimeoutMs	number	1800000	HTTP timeout (default 30 min)
debug	boolean	false	Enable debug logging
stateExtractor	function	undefined	Custom state extraction

createIsolatedAgent Options

Option	Type	Default	Description
deploymentUrl	string	required	LangGraph deployment URL
graphId	string	required	Graph identifier
langsmithApiKey	string	undefined	LangSmith API key
clientTimeoutMs	number	1800000	HTTP timeout
debug	boolean	false	Enable debug mode

Advanced Usage

Custom State Extraction

If you need to extract custom fields from the CopilotKit request:

const runner = new HistoryHydratingAgentRunner({
  agent,
  deploymentUrl,
  graphId,
  stateExtractor: (input, forwardedProps) => ({
    // Extract from forwardedProps.configurable (useCoAgent config)
    tenantId: forwardedProps?.configurable?.tenantId as string,
    userId: forwardedProps?.configurable?.userId as string,
    // Or from input.state (useCoAgent initialState)
    ...input.state,
  }),
});

Why createIsolatedAgent?

In serverless environments (especially Vercel Fluid Compute), Node.js module-level state can be shared between bundled routes. This causes a critical bug where the LangGraph deployment URL gets contaminated between different agent configurations.

createIsolatedAgent fixes this by:

1. Creating agents with frozen, immutable config
2. Verifying the internal client URL matches expected
3. Force-replacing the client if contamination is detected

Always use createIsolatedAgent instead of new LangGraphAgent() in serverless environments.

Debug Mode

Enable debug logging to troubleshoot issues:

const runner = new HistoryHydratingAgentRunner({
  // ...
  debug: true,
});

This logs:

• History fetching progress
• Message transformation details
• Stream processing events
• State extraction results

How It Works

History Hydration Flow

When a client connects to an existing thread:

1. Fetch History: Retrieves all checkpoints from LangGraph via client.threads.getHistory()
2. Extract Messages: Processes checkpoints chronologically, deduplicating messages by ID
3. Transform Format: Converts LangGraph messages to CopilotKit format
4. Emit Events: Sends MESSAGES_SNAPSHOT and STATE_SNAPSHOT events to frontend
5. Join Stream: If thread is busy, joins the active execution stream

Event Types Handled

• on_chat_model_stream → TEXT_MESSAGE_CONTENT
• on_chat_model_start → TEXT_MESSAGE_START
• on_chat_model_end → TEXT_MESSAGE_END
• on_tool_start → TOOL_CALL_START
• on_tool_end → TOOL_CALL_END
• Custom CopilotKit events (manual message/tool/state emission)
• Interrupt events

API Reference

Exports

// Core
export { HistoryHydratingAgentRunner } from "copilotkit-langgraph-history";
export { createIsolatedAgent } from "copilotkit-langgraph-history";

// Types
export type {
  HistoryHydratingRunnerConfig,
  StateExtractor,
  CreateIsolatedAgentConfig,
  LangGraphMessage,
  ThreadState,
} from "copilotkit-langgraph-history";

// Constants
export {
  DEFAULT_TIMEOUT,
  DEFAULT_HISTORY_LIMIT,
  MAX_HISTORY_LIMIT,
} from "copilotkit-langgraph-history";

// Event Enums
export {
  CustomEventNames,
  LangGraphEventTypes,
} from "copilotkit-langgraph-history";

// Utilities (advanced)
export {
  transformMessages,
  extractContent,
  processStreamChunk,
} from "copilotkit-langgraph-history";

Environment Variables

# Required
LANGGRAPH_DEPLOYMENT_URL=https://your-deployment.langchain.com

# Optional (for authentication)
LANGSMITH_API_KEY=your-api-key

Troubleshooting

"No history found for thread"

• Ensure the thread exists in LangGraph
• Check that deploymentUrl is correct
• Verify langsmithApiKey has access to the deployment

Messages not loading on refresh

• Confirm threadId is being passed to <CopilotKit>
• Check browser console for hydration errors
• Enable debug: true to see detailed logs

"URL mismatch detected" warning

This is expected when the runner detects and fixes serverless state contamination. The client is automatically replaced with the correct URL.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

MIT License - see LICENSE for details.

Credits

Created by Daniel Frey.

Inspired by the need for thread history persistence in CopilotKit + LangGraph applications.