What happens to my LangChain response if Monetzly is down?

The SDK falls back to yielding your original token stream, so the user still receives a complete LangChain answer — just without ads for that request.

Do I need a manual fallback in LangChain?

The SDK handles fallback automatically, but wrapping the loop in your usual error handling is still good practice.

How-to guide

How to Handle Ad-Injection Errors in LangChain

This guide shows how to handle ad-injection errors in a LangChain app, the goal being to make sure the user still gets a reply if the ad service fails. It builds on the Monetzly server SDK, so the approach is specific to how LangChain produces and streams responses.

Overview

Monetization should never break your LangChain app. The Monetzly SDK is built for this: if the ad service connection fails or errors mid-stream, it falls back to yielding your original, un-injected token stream, so the user still gets a complete answer.

You should still wrap the injection loop in your normal error handling so a failure degrades to a plain LangChain response rather than a broken request.

LangChain apps are usually built for RAG chatbots, tool-using agents, multi-step reasoning pipelines, and document QA, so handle ad-injection errors typically comes up while a user is mid-conversation, the moment where monetization has to be additive rather than disruptive.

How this works on LangChain

Grounded in this repo (src/app/api/agents/chat/route.ts) and the SDK's own LangChain example. A LangChain.js chat model returns a stream of message chunks that already carry a `content` field, so it can be passed straight into sdk.inject() with no adapter.

Because Monetzly's inject() accepts any async token stream, the LangChain side of this task is just mapping your output to it, no rewrite of your LangChain model call.

Steps

Wrap the injection loop in your standard LangChain error handling.
Rely on the SDK's automatic fallback to the original stream on ad-service failure.
Log injection failures so you can monitor fill and uptime separately from your model.

Integration snippet

import { MonetzlySDK, MonetzlyConfig } from "@monetzly/server-sdk";
import { ChatGoogleGenerativeAI } from "@langchain/google-genai";

const config: MonetzlyConfig = {
  apiKey: process.env.MONETZLY_API_KEY!,
  serverAddress: process.env.MONETZLY_SERVER_ADDRESS || "localhost:8080",
  useSSL: !process.env.MONETZLY_SERVER_ADDRESS?.startsWith("localhost"),
};
const sdk = new MonetzlySDK(config);
const llm = new ChatGoogleGenerativeAI({ model: "gemini-2.0-flash" });

await sdk.connect();
const llmStream = await llm.stream(messages); // chunks already have .content

for await (const token of sdk.inject(llmStream, {
  sessionId,
  prompt: messages[messages.length - 1]?.content ?? "",
})) {
  send(token.content ?? ""); // forward to client (SSE / WebSocket)
}
await sdk.disconnect();