Gemini Adapter

The Gemini adapter integrates with Google's @google/generative-ai SDK. It translates Tekimax's chat format into Gemini's session-based startChat() / sendMessage() model.

Installation

Code

npm install tekimax-ts

Usage

Code

import { Tekimax, GeminiProvider } from 'tekimax-ts';

const client = new Tekimax({
  provider: new GeminiProvider({
    apiKey: process.env.GOOGLE_API_KEY!,
  })
});

const result = await client.text.chat.completions.create({
  model: 'gemini-1.5-pro',
  messages: [{ role: 'user', content: 'Hello Gemini' }]
});

console.log(result.message.content);

Streaming

Code

const stream = client.text.chat.completions.createStream({
  model: 'gemini-1.5-flash',
  messages: [{ role: 'user', content: 'List 5 fun facts about space' }]
});

for await (const chunk of stream) {
  process.stdout.write(chunk.delta);
}

Tool Calling

The adapter maps the standard tools array to Gemini's functionDeclarations format. Gemini returns functionCall objects in its response, which the adapter normalizes to the unified toolCalls array.

Code

const result = await client.text.chat.completions.create({
  model: 'gemini-1.5-pro',
  messages: [{ role: 'user', content: 'What is the weather in Tokyo?' }],
  tools: [{
    type: 'function',
    function: {
      name: 'get_weather',
      description: 'Get current weather',
      parameters: { type: 'object', properties: { location: { type: 'string' } } }
    }
  }]
});

console.log(result.message.toolCalls);

Image Analysis (Vision)

Gemini supports image analysis via inlineData. The adapter auto-fetches remote URLs and converts them to base64 — Gemini's API does not accept remote image URLs directly.

Code

const analysis = await client.images.analyze({
  model: 'gemini-1.5-flash',
  image: 'https://example.com/chart.png',
  prompt: 'Extract the data from this chart.'
});

console.log(analysis.content);

Video Analysis

Gemini is the only provider with native video understanding. The adapter downloads remote video URLs and converts them to inlineData automatically.

Code

const analysis = await client.videos.analyze({
  model: 'gemini-1.5-flash', // Flash has a 1M token context window at lower cost
  video: 'https://cdn.example.com/demo.mp4',
  prompt: 'Summarize what happens in this video.'
});

console.log(analysis.content);

Notes

Session Model: Gemini uses startChat() + sendMessage() internally. The adapter creates a session from your message history (all messages except the last) and sends the last message as the input. This preserves multi-turn context.
Model Names: Use the full model identifier: gemini-1.5-pro, gemini-1.5-flash, gemini-pro. Note that gemini-pro maps to the older Gemini 1.0 Pro.
Image URLs: Gemini doesn't accept remote URLs directly in the inlineData format. The adapter fetches the image, converts it to base64, and detects the MIME type automatically. For best performance, pass base64 data URIs directly if you already have the image data.

Gemini Adapter

Installation

Usage

Streaming

Tool Calling

Image Analysis (Vision)

Video Analysis

Notes

On this page