How to Handle Ad-Injection Errors in OpenAI Assistants API

Overview

Monetization should never break your OpenAI Assistants API 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 OpenAI Assistants API response rather than a broken request.

OpenAI Assistants API apps are usually built for hosted chat assistants, file-search agents, and tool-calling assistants, 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 OpenAI Assistants API

The Node OpenAI SDK streams run/completion events. Map each text delta event to { content: delta } and feed the generator into sdk.inject(). The Monetzly side is standard; the only framework detail to confirm is the exact delta event field for your chosen mode (Assistants run stream vs Chat Completions).

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

Steps

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

Integration snippet

import OpenAI from "openai";
import { MonetzlySDK } from "@monetzly/server-sdk";

const openai = new OpenAI();
const sdk = new MonetzlySDK({
  apiKey: process.env.MONETZLY_API_KEY!,
  serverAddress: process.env.MONETZLY_SERVER_ADDRESS!,
});
await sdk.connect();

const stream = await openai.chat.completions.create({
  model: "gpt-4o-mini", messages, stream: true,
});

async function* asChunks() {
  // TODO_VERIFY: for the Assistants run stream, read the text delta from the
  // run event instead of choices[0].delta.content.
  for await (const ev of stream) yield { content: ev.choices[0]?.delta?.content ?? "" };
}

for await (const t of sdk.inject(asChunks(), { prompt })) send(t.content ?? "");
await sdk.disconnect();

Frequently asked questions

Related