feat(examples): replace langgraph-chat with deepagents-based langchain-chat

[christian-bromann]

What

Updating the LangChain example to use stream transformer to simplify user code. I would love to see OpenUI maintain some of these primitives, e.g. examples/langchain-chat/src/lib as well as the stream transformer, in so users wouldn't need to maintain the code themselves. Ideally all the user needs to do to use LangChain with OpenUI is:

Define an agent, e.g.:

import { streamTransformer } from "@openuidev/langchain";

export const graph = createDeepAgent({
  model: `openai:gpt-5.5`,
  tools: [getWeather, getStockPrice, searchWeb],
  systemPrompt: SYSTEM_PROMPT,
  streamTransformers: [openUIStreamTransformer],
});

Define the custom API endpoint in Next.js:

import { createLangChainStreamResponse } from "@openuidev/langchain";
import { NextRequest } from "next/server";

export const runtime = "nodejs";

/**
 * Local LangGraph server configuration defined via langgraph.json
 */
const API_URL = process.env.LANGGRAPH_API_URL || "http://localhost:2024";
const ASSISTANT_ID = process.env.LANGGRAPH_ASSISTANT_ID || "agent";

/**
 * Proxies the browser <-> LangGraph server. The browser posts native AG-UI
 * messages; {@link createLangChainStreamResponse} converts them to LangChain
 * messages, starts the graph over the protocol-v2 endpoints, and relays only
 * the custom OpenUI channel.
 */
export async function POST(req: NextRequest) {
  return createLangChainStreamResponse(req, {
    apiUrl: API_URL,
    assistantId: ASSISTANT_ID,
  });
}

The stream transformer transforms the agent stream to be AGUI compatible so that in the frontend code, no LangChain specific adapter is needed.

Changes

• Renamed/rewrote examples/langgraph-chat to examples/langchain-chat, swapping the hand-built supervisor graph for a single createDeepAgent agent (src/agent/agent.ts) that loads the generated OpenUI system prompt via readFileSync (the LangGraph dev server's tsx/CJS loader can't handle with { type: "text" } import attributes).
• Added a local openUIStreamTransformer (src/agent/openui-transformer.ts) that consumes @langchain/protocol MessagesData events and emits AG-UI events on a StreamChannel.remote("openui") channel, with no namespace filtering so nested model_request:*/tools:* messages are forwarded.
• Added src/lib/stream-openui.ts, a framework-agnostic helper that drives an agent-protocol-v2 run, subscribes to custom:openui (no namespace/depth filter), unwraps the forwarded AG-UI events, and relays them as SSE — fully TSDoc-documented.
• Extracted the request→stream→Response glue into createLangChainStreamResponse (src/lib/langchain-stream-response.ts) so OpenUI can later export it as a reusable helper; the route handler (src/app/api/chat/route.ts) is now a thin delegator.
• Simplified the frontend (src/app/page.tsx) to use OpenUI's default apiUrl adapter (removed the custom processMessage/streamProtocol wiring) and dropped the local LANGSMITH_API_KEY read since this is a local demo (the apiKey option remains optional on the helpers).
• Used the web-standard global crypto.randomUUID() instead of importing from node:crypto.
• Updated package.json (deepagents/langchain deps, dev script that runs the UI and LangGraph server together), README.md, and .env.example to match the new architecture; refreshed pnpm-lock.yaml.

Test Plan

Describe how you validated this change.

• Not applicable (explain why)
• Verified locally

Checklist

• I linked a related issue, if applicable
• I updated docs/README when needed
• I considered backwards compatibility