Recipe — Use OpenAI through the Gateway

This recipe walks through standing up an AI Wrapper project so existing OpenAI-using apps can route through PromptGate without code changes.

End state: your apps call <your-gateway>/api/<uuid>/v1/chat/completions with the OpenAI SDK unchanged, and you get token-gated access, observability, rate limits, budgets, and the option to swap to a different provider later.

Prerequisites

• PromptGate installed and running (Installation)
• An OpenAI API key (sk-…)
• Logged into the admin UI

Step 1 — Add the OpenAI credential

Top-right user menu → Credentials → + New credential:

• Name: OpenAI Production
• Provider: OpenAI
• Secret: paste your sk-…

Save. The key is encrypted with APP_KEY at rest.

Step 2 — Create an AI Wrapper project

Sidebar → Projects → + New project:

• Name: OpenAI Proxy
• Type: AI Wrapper
• Environment: prod

Click your new project to enter it.

Step 3 — Bind OpenAI to the wrapper

Project sidebar → Providers:

• Tick Enabled on the OpenAI row.
• Pick Credential → OpenAI Production.

Save.

(You can do this for as many providers as you want — each one bound to its credential. The wrapper will route requests by provider:model accordingly.)

Step 4 — (Optional) Add a friendly alias

Project sidebar → Aliases → + New alias:

• Alias: fast
• Provider: openai
• Model: gpt-4o-mini

Now clients can use model: "fast" instead of model: "openai:gpt-4o-mini". Swap the alias’s underlying model later without changing client code.

Step 5 — Issue a token

Project sidebar → API Tokens → + New token:

• Name: Production app
• Environment: live
• Scopes: chat

Save. Copy the plaintext (pg_live_…) — it’s only shown once.

Step 6 — Point the OpenAI SDK at PromptGate

The wrapper’s URL is your gateway URL + /api/<UUID>/v1:

PG_BASE=https://gateway.your-domain.com/api/8e3f-…/v1

Python

import os
from openai import OpenAI

client = OpenAI(
    base_url=os.environ['PG_BASE'],          # PromptGate's wrapper URL
    api_key=os.environ['PG_TOKEN'],          # The pg_live_... token
)

resp = client.chat.completions.create(
    model="promptgate:fast",                  # Or "openai:gpt-4o-mini" / "openai:gpt-4o"
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

Node.js

import OpenAI from 'openai';

const client = new OpenAI({
    baseURL: process.env.PG_BASE,
    apiKey: process.env.PG_TOKEN,
});

const resp = await client.chat.completions.create({
    model: 'promptgate:fast',
    messages: [{ role: 'user', content: 'Hello!' }],
});
console.log(resp.choices[0].message.content);

LangChain

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url=os.environ['PG_BASE'],
    api_key=os.environ['PG_TOKEN'],
    model="promptgate:fast",
)
print(llm.invoke("Hello!").content)

Vercel AI SDK

import { createOpenAI } from '@ai-sdk/openai';
import { generateText } from 'ai';

const promptgate = createOpenAI({
    baseURL: process.env.PG_BASE,
    apiKey: process.env.PG_TOKEN,
});

const { text } = await generateText({
    model: promptgate('promptgate:fast'),
    prompt: 'Hello!',
});

Step 7 — Verify in the UI

Trigger a request, then watch:

• Live Logs — request appears within 2s with status, latency, tokens.
• Metrics — KPIs and charts update.
• Audit Log — auth.token.used events.

What you’ve gained

• ✅ Provider key out of every app — replace it once in PromptGate.
• ✅ Token-gated access — issue per-app tokens, revoke independently.
• ✅ Observability — every request logged, rolled up, exportable.
• ✅ Rate limits + budgets available (configure on the alias level via AI Endpoints for fine grain, or on the wrapper-wide via roadmap).
• ✅ Provider portability — swap fast from openai:gpt-4o-mini to groq:llama-3.1-8b-instant in 5 seconds. Clients never know.

What’s next

• Add a guardrail (PII filter mask mode) so customer data never leaves your gateway: Guardrails.
• Add rate limits per minute / per hour: Rate Limits.
• Add monthly budget: Budgets.
• Add Anthropic + Mistral as additional aliases for cheap fallback.

---

Next: Multi-provider AI Wrapper.

---

© Akyros Labs LLC. All rights reserved.