Multimodal AI Gateway

Supports Gemini (Google Gen AI SDK; requests use /v1beta paths) and OpenAI / DeepSeek (OpenAI SDK; baseURL includes /v1). Set SDK baseUrl to https://<host>/g/<public-id> (do not append /v1beta). The model field in each request works like the official SDK; the gateway endpoint resolves it using console logical routes and model connections.

Platform API key

Requests must use a platform API key from the console (not your upstream Gemini/OpenAI key), sent as x-goog-api-key (Gemini) or Authorization: Bearer (OpenAI). The key must belong to the Gateway’s team. If the key was created bound to a Gateway, it may only call that Gateway’s /g/<public-id>/... paths. Unbound keys may call any Gateway in the same team (the public-id in the URL must still match). Binding the key to the target Gateway in API keys is recommended.

Create keys under API keys; configure upstream provider keys (model connections) on the Gateway detail page.

Prerequisites

Public ID: the 10-character code after /g/ on the Gateway detail page → Gateway (upstream API keys required)
Platform API key: see API keys (bind to target Gateway when possible)
Logical routes & model connections: configure in the Gateway console; the gateway resolves the request model at the endpoint (only one route stays enabled per Gateway when you save)

baseUrl = "https://ai-sass-7pymefszf-keyskulls-projects.vercel.app/g/<public-id>"   # do not append /v1beta

Access protocols

Enable client entry points under Access protocols on the Gateway detail page. Disabled protocols return 403. All built-in protocols are enabled by default.

Protocol id	Path (under /g/<public-id>/)	Client
gemini-v1beta	v1beta/**	Google Gen AI SDK (baseUrl without /v1beta)
openai-chat	v1/chat/completions	OpenAI SDK · Chat Completions (baseURL includes /v1)
openai-responses	v1/responses	OpenAI SDK · Responses API

Gemini SDK also uses /upload/v1beta/... for uploads; no separate protocol toggle (shares baseUrl with Gemini).

JavaScript (@google/genai)

import { GoogleGenAI } from "@google/genai";

const ai = new GoogleGenAI({
  apiKey: process.env.ENGRA_API_KEY!,
  httpOptions: {
    baseUrl: "https://ai-sass-7pymefszf-keyskulls-projects.vercel.app/g/<public-id>",
  },
});

// model comes from the request, same as the official SDK
await ai.models.generateContent({
  model: "your-model",
  contents: "Hello",
});

Python (google-genai)

import os
from google import genai

client = genai.Client(
    api_key=os.environ["ENGRA_API_KEY"],
    http_options={"base_url": "https://ai-sass-7pymefszf-keyskulls-projects.vercel.app/g/<public-id>"},
)

# model from request, same as the official SDK
client.models.generate_content(model="your-model", contents="Hello")

OpenAI / DeepSeek

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.ENGRA_API_KEY!,
  baseURL: "https://ai-sass-7pymefszf-keyskulls-projects.vercel.app/g/<public-id>/v1",
});

await client.chat.completions.create({
  model: "your-model",  // model from request
  messages: [{ role: "user", content: "Hello" }],
});

Requires openai-chat. Configure model connections on the Gateway for Gemini, OpenAI, DeepSeek, and more.

curl example (Gemini generateContent)

Replace YOUR_PLATFORM_KEY and public-id with values from the console.

curl "https://ai-sass-7pymefszf-keyskulls-projects.vercel.app/g/<public-id>/v1beta/models/your-model:generateContent" \
  -H "x-goog-api-key: YOUR_PLATFORM_KEY" \
  -H "Content-Type: application/json" \
  -d '{"contents":[{"parts":[{"text":"Hello"}]}]}'

Common errors

HTTP	Typical cause
401	Invalid or missing platform API key (Gemini: x-goog-api-key; OpenAI: Authorization header)
403	Access protocol disabled (e.g. gemini-v1beta, openai-chat)
404	Invalid public_id or Gateway not enabled
503	No upstream API key configured under model connections

Configure logical routes

Configure logical routes, model connections, IF/ELSE conditions (input format or headers/metadata), fallback chains, prompt templates, temperature/JSON/embedding dimensions, etc. Clients pass model in each request as with the upstream SDK; the gateway endpoint resolves it.

→ Logical routes guide (with canvas demo)→ Console