How to Test Ad Injection Locally in OpenAI Assistants API

Overview

You can develop ad injection for a OpenAI Assistants API app locally before going to production. Point the SDK at a local ad-server address and turn SSL off; everything else — the injection call, session context, fallback — behaves the same as production.

This lets you verify that your OpenAI Assistants API stream flows through injection and reaches the client correctly without needing production credentials wired up.

OpenAI Assistants API apps are usually built for hosted chat assistants, file-search agents, and tool-calling assistants, so test ad injection locally 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. Set the server address to your local host (e.g. localhost:8080) and disable SSL.
2. Run your OpenAI Assistants API app and send a test prompt through the injection path.
3. Confirm tokens stream to the client, then switch to your production host and SSL for release.

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