Gemini and Vertex AI

IntelliNode reaches Gemini in two places: the Gemini Developer API and Vertex AI on Google Cloud. The same code works for both, and the key you have decides which one you use.

Which key you have

Your key	Provider	What you get
An AI Studio key from aistudio.google.com	'gemini'	The Gemini Developer API
A Google Cloud Agent Platform key (it starts with AQ.)	'vertex'	Vertex AI in express mode
The same key plus a projectId	'vertex'	Project mode, which adds Veo video and the Live API on Vertex AI

A Vertex key sent to the Developer API fails with API_KEY_SERVICE_BLOCKED, so match the provider to the key. With no key at all, a projectId uses your gcloud login (Application Default Credentials) in Node.

Chat

Use Chatbot as with any other provider:

const { Chatbot } = require('intellinode');

const bot = new Chatbot(process.env.VERTEX_API_KEY, 'vertex');   // or (GEMINI_API_KEY, 'gemini')
const input = Chatbot.createInput('vertex', 'You are concise.', {
  systemInstruction: true,
  generationConfig: { thinkingConfig: { thinkingLevel: 'low' } },
});
input.addUserMessage('Describe this image.', [{ data: base64Png, mimeType: 'image/png' }]);

const [answer] = await bot.chat(input);
for await (const chunk of bot.stream(input)) process.stdout.write(chunk);

Two options are worth setting:

• systemInstruction: true sends your system message as a real Gemini system instruction.
• Gemini 3 models think before they answer, and the thinking counts toward maxTokens. A low thinking level keeps chat fast. If you get an empty answer, raise maxTokens.

For project mode, pass options as the fourth argument: new Chatbot(key, 'vertex', null, { projectId, location: 'global' }). Gen takes the same provider ids, for example Gen.generate_text('Write a haiku', key, 'vertex').

Everything else: GoogleAIWrapper

GoogleAIWrapper covers the features only Gemini has:

const fs = require('fs');
const { GoogleAIWrapper } = require('intellinode');

const google = new GoogleAIWrapper(process.env.VERTEX_API_KEY, { vertex: true });   // or new GoogleAIWrapper(GEMINI_API_KEY)

// answer from Google Search, with sources
const grounded = await google.generateContent({ contents: 'Latest Node.js LTS?', tools: [{ googleSearch: {} }] });
console.log(GoogleAIWrapper.extractText(grounded));
console.log(GoogleAIWrapper.extractCitations(grounded));   // [{ title, uri, domain }]

// an image
const [image] = GoogleAIWrapper.extractImages(await google.generateImage('A watercolor fox', { imageConfig: { aspectRatio: '16:9' } }));
fs.writeFileSync('fox.png', Buffer.from(image.data, 'base64'));

// speech, as a WAV buffer
fs.writeFileSync('hello.wav', await google.textToSpeech('Welcome to IntelliNode', { voice: 'Puck' }));

// read a PDF, image, audio or video file
const summary = await google.mediaToText('Summarize this report.', ['report.pdf']);

startChat() keeps the history for you, which Gemini 3 needs for function calling:

const chat = google.startChat({ systemInstruction: 'You are a tutor.' });
await chat.sendText('My name is Sam.');
for await (const chunk of chat.stream('What is my name?')) process.stdout.write(chunk);

Save chat.history as JSON and resume later with startChat({ history }).

Video, music and voice

Video and music are billed per call, and a video takes minutes. Music works with either key. Veo video runs on the Developer API, or on Vertex AI in project mode ({ vertex: true, projectId }):

const project = new GoogleAIWrapper(process.env.VERTEX_API_KEY, { vertex: true, projectId });

const operation = await project.generateVideo('Drone shot over a misty forest', { durationSeconds: 4, resolution: '720p' });
const [video] = GoogleAIWrapper.extractVideos(await project.waitForVideoCompletion(operation));

const [track] = GoogleAIWrapper.extractAudio(await google.generateMusic('Calm piano with soft rain'));
fs.writeFileSync('music.wav', GoogleAIWrapper.audioToWav(track));

The Live API (liveConnect, liveGenerate) holds a voice session over a WebSocket, with the same rule as video on Vertex AI. It needs Node 22, or node --experimental-websocket on Node 20.

What works where

Feature	Developer API	Vertex express	Vertex project
Chat, streaming, JSON, function calling	yes	yes	yes
Google Search grounding	yes	yes	yes
Images, speech, music, embeddings, media understanding	yes	yes	yes
Veo video	yes	no	yes
Live API	yes	no	yes
Files API	yes	no	no
Context caching	yes	no	with OAuth

Errors come back as GoogleAIError with status and details, and IntelliNode removes your key from the message. The Assistant uses all of this to build a Gemini-style chat app.