Back to AI Framework Overview | All Providers

@memberjunction/ai-gemini

MemberJunction AI provider for Google Gemini models. Provides both LLM and image generation capabilities, supporting Gemini 2.5, Gemini 3, and Flash model families with native multimodal support, thinking/reasoning, and streaming.

Architecture

graph TD
    A["GeminiLLM
(Chat Provider)"] -->|extends| B["BaseLLM
(@memberjunction/ai)"]
    C["GeminiImageGenerator
(Image Provider)"] -->|extends| D["BaseImageGenerator
(@memberjunction/ai)"]
    A -->|wraps| E["GoogleGenAI
(@google/genai)"]
    C -->|wraps| E
    A -->|provides| F["Chat + Streaming"]
    A -->|provides| G["Thinking/Reasoning
Budget Control"]
    C -->|provides| H["Image Generation
+ Editing + Variations"]
    B -->|registered via| I["@RegisterClass"]
    D -->|registered via| I

    style A fill:#7c5295,stroke:#563a6b,color:#fff
    style C fill:#7c5295,stroke:#563a6b,color:#fff
    style B fill:#2d6a9f,stroke:#1a4971,color:#fff
    style D fill:#2d6a9f,stroke:#1a4971,color:#fff
    style E fill:#2d8659,stroke:#1a5c3a,color:#fff
    style F fill:#b8762f,stroke:#8a5722,color:#fff
    style G fill:#b8762f,stroke:#8a5722,color:#fff
    style H fill:#b8762f,stroke:#8a5722,color:#fff
    style I fill:#b8762f,stroke:#8a5722,color:#fff
Copy

Features

LLM (GeminiLLM)

• Chat Completions: Full conversational AI with system instructions
• Streaming: Real-time response streaming with chunk processing
• Thinking/Reasoning: Configurable thinking budget for Gemini 2.5+ and thinking levels for Gemini 3+
• Multimodal Input: Native support for text, images, audio, video, and file inputs
• Message Alternation: Automatic handling of Gemini's role alternation requirements
• Safety Handling: Detection and reporting of content blocking with detailed safety ratings
• Effort Level Mapping: Maps MJ effort levels (1-100) to Gemini thinking budgets (0-24576)

Image Generation (GeminiImageGenerator)

• Text-to-Image: Generate images using Gemini image models
• Image Editing: Edit existing images using multimodal context
• Image Variations: Create variations of existing images
• Resolution Control: Support for sizes up to 4K (3840x2160)
• Style and Quality: Configurable style and quality parameters

Embeddings (GeminiEmbedding)

• Multimodal Embeddings: Map text, images, video, audio, and PDF into one shared 3072-dimensional vector space
• Cross-Modal Retrieval: Embed text and media together so a text query can match an image, audio, or video
• Text and Batch: Single and batch text embedding

Installation

npm install @memberjunction/ai-gemini
Copy

Usage

Chat Completion

import { GeminiLLM } from "@memberjunction/ai-gemini";

const llm = new GeminiLLM("your-google-api-key");

const result = await llm.ChatCompletion({
    model: "gemini-2.5-flash",
    messages: [
        { role: "system", content: "You are a helpful assistant." },
        { role: "user", content: "Explain quantum computing." },
    ],
    temperature: 0.7,
});
Copy

Streaming with Thinking

const result = await llm.ChatCompletion({
    model: "gemini-2.5-pro",
    messages: [{ role: "user", content: "Solve this math problem step by step." }],
    effortLevel: "75",
    streaming: true,
    streamingCallbacks: {
        OnContent: (content) => process.stdout.write(content),
    },
});

console.log("Thinking:", result.data.choices[0].message.thinking);
Copy

Image Generation

import { GeminiImageGenerator } from "@memberjunction/ai-gemini";

const generator = new GeminiImageGenerator("your-google-api-key");

const result = await generator.GenerateImage({
    prompt: "A futuristic city at night",
    model: "gemini-3-pro-image-preview",
    size: "2048x2048",
});
Copy

Embeddings

import { GeminiEmbedding } from "@memberjunction/ai-gemini";

const embedding = new GeminiEmbedding("your-google-api-key");

// Text (3072-dim vector)
const text = await embedding.EmbedText({ text: "a golden retriever in the snow" });

// Multimodal: text + image fused into ONE vector (cross-modal retrieval)
const multimodal = await embedding.EmbedContent({
    content: [
        { type: "text", content: "product photo:" },
        { type: "image_url", content: "<base64-image>", mimeType: "image/png" },
    ],
});
console.log(multimodal.vector.length); // 3072
Copy

Thinking Budget / Effort Level

The provider maps MJ effort levels to Gemini's thinking system:

Supported Parameters

Class Registration

• GeminiLLM -- Registered via @RegisterClass(BaseLLM, 'GeminiLLM')
• GeminiImageGenerator -- Registered via @RegisterClass(BaseImageGenerator, 'GeminiImageGenerator')
• GeminiEmbedding -- Registered via @RegisterClass(BaseEmbeddings, 'GeminiEmbedding')

Dependencies

• @memberjunction/ai - Core AI abstractions
• @memberjunction/global - Class registration
• @google/genai - Google GenAI SDK