gemini-rs

A Rust wrapper for the Google Cloud Gemini REST API (Vertex AI).

Features

Text Generation: Generate text responses from Gemini models
Streaming: Stream responses in real-time using SSE
Text Embeddings: Generate text embeddings for ML pipelines
Token Counting: Count tokens before making API calls
Image Generation: Generate and predict images
Conversations: Maintain multi-turn dialogue context
Safety Settings: Configure content safety thresholds
Function Calling: Declare and use tools/functions
Google Search Retrieval: Enable dynamic search-backed responses
JSON Schema Responses: Get structured JSON output

Installation

Add this to your Cargo.toml:

[dependencies]
gemini-rs = { path = "path/to/gemini-rs" }
tokio = { version = "1", features = ["full"] }

Quick Start

use gemini_rs::prelude::*;
use gcp_auth::AuthenticationManager;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Authenticate with GCP
    let provider = AuthenticationManager::new();
    
    // Create client
    let client = GeminiClient::new(
        provider,
        "us-central1-aiplatform.googleapis.com".to_string(),
        "your-project-id".to_string(),
        "us-central1".to_string(),
    );
    
    // Generate content
    let request = GenerateContentRequest::builder()
        .contents(vec![Content {
            role: Some(Role::User),
            parts: Some(vec![Part::Text("Hello, Gemini!".to_string())]),
        }])
        .build();
    
    let response = client.generate_content(&request, "gemini-2.0-flash").await?;
    
    for candidate in &response.candidates {
        if let Some(text) = candidate.get_text() {
            println!("{}", text);
        }
    }
    
    Ok(())
}

Examples

Run the examples in the examples/ directory:

# Generate text
cargo run --example text-from-text

# Streaming text generation
cargo run --example text-from-text-streaming

# Multi-turn conversation
cargo run --example conversation

# Count tokens
cargo run --example count-tokens

# Generate images
cargo run --example generate_image

# Text embeddings
cargo run --example text-embedding

# JSON schema responses
cargo run --example json-schema

# Safety settings
cargo run --example safety-setting

# System instructions
cargo run --example system_instruction

# Google Search Retrieval
cargo run --example google-search-retrieval

# Google Search tool
cargo run --example google-search

API Reference

Client Methods

Method	Description
`generate_content()`	Generate content from a model
`generate_content_stream()`	Stream content generation responses
`stream_generate_content()`	Stream content using a queue-based approach
`prompt_conversation()`	Send a conversation prompt and get response
`text_embeddings()`	Generate text embeddings
`count_tokens()`	Count tokens in a request
`predict_image()`	Generate/predict images

Supported Models

The library works with all Vertex AI Gemini models, including:

gemini-2.0-flash
gemini-2.0-flash-lite
gemini-1.5-pro
gemini-1.5-flash
text-embedding-004
imagen-3.0-generate-002

License

See LICENSE for details.