llm-connector

Next-generation Rust library for LLM protocol abstraction with native multi-modal support.

Supports 12+ providers: OpenAI, Anthropic, Google, Aliyun, Zhipu, Ollama, Tencent, Volcengine, DeepSeek, Moonshot, Xiaomi, and more.

Key Features

• 12+ Provider Support - OpenAI, Anthropic, Google, Aliyun, Zhipu, Ollama, Tencent, Volcengine, DeepSeek, Moonshot, Xiaomi
• Unified Output Format - All providers return the same StreamingResponse type
• Multi-modal Support - Text + images in a single message (details)
• Function Calling - OpenAI-compatible tools with streaming (details)
• Reasoning Models - Universal support for reasoning models (details)
• Real-time Streaming - Universal streaming with format abstraction (details)
• Extreme Performance - 7,000x+ faster client creation (7µs vs 53ms)
• Type-Safe - Full Rust type safety with Result-based error handling

Installation

[dependencies]
llm-connector = "0.6.0"
tokio = { version = "1", features = ["full"] }

# Streaming is included by default since v0.6.0
# To disable default features:
# llm-connector = { version = "0.6.0", default-features = false, features = ["reqwest"] }

Quick Start

use llm_connector::{LlmClient, types::{ChatRequest, Message, Role}};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = LlmClient::openai("sk-...")?;

    let request = ChatRequest {
        model: "gpt-4".to_string(),
        messages: vec![Message::text(Role::User, "Hello!")],
        ..Default::default()
    };

    let response = client.chat(&request).await?;
    println!("Response: {}", response.content);
    Ok(())
}

Supported Providers

📖 Detailed documentation: docs/PROVIDERS.md

Builder Pattern

use llm_connector::LlmClient;

// Simple
let client = LlmClient::builder()
    .deepseek("sk-...")
    .build()?;

// With custom configuration
let client = LlmClient::builder()
    .openai("sk-...")
    .base_url("https://api.deepseek.com")
    .timeout(60)
    .proxy("http://proxy:8080")
    .build()?;

// Ollama (no API key needed)
let client = LlmClient::builder()
    .ollama()
    .base_url("http://192.168.1.100:11434")
    .build()?;

Streaming

use llm_connector::{LlmClient, types::{ChatRequest, Message, Role}};
use futures_util::StreamExt;

let client = LlmClient::openai("sk-...")?;
let request = ChatRequest {
    model: "gpt-4".to_string(),
    messages: vec![Message::text(Role::User, "Tell me a story")],
    stream: Some(true),
    ..Default::default()
};

let mut stream = client.chat_stream(&request).await?;
while let Some(chunk) = stream.next().await {
    if let Some(content) = chunk?.get_content() {
        print!("{}", content);
    }
}

All providers return the same StreamingResponse type. 📖 Streaming Guide

Function Calling / Tools

use llm_connector::{LlmClient, ChatRequest, Message, Tool, ToolChoice};

let tools = vec![Tool::function(
    "get_weather",
    Some("Get weather info".to_string()),
    serde_json::json!({
        "type": "object",
        "properties": { "location": { "type": "string" } },
        "required": ["location"]
    }),
)];

let request = ChatRequest::new("gpt-4")
    .add_message(Message::user("What's the weather in Beijing?"))
    .with_tools(tools)
    .with_tool_choice(ToolChoice::auto());

let response = client.chat(&request).await?;

// Convenience methods for tool call handling
if response.is_tool_call() {
    for tc in response.tool_calls() {
        println!("Call: {} args: {}", tc.function.name, tc.function.arguments);
        let args: serde_json::Value = tc.parse_arguments()?;
    }
}

📖 Detailed documentation: docs/TOOLS.md

Multi-modal Content

use llm_connector::{LlmClient, types::{ChatRequest, Message, MessageBlock, Role}};

let request = ChatRequest {
    model: "gpt-4o".to_string(),
    messages: vec![
        Message::new(
            Role::User,
            vec![
                MessageBlock::text("What's in this image?"),
                MessageBlock::image_url("https://example.com/image.jpg"),
            ],
        ),
    ],
    ..Default::default()
};

📖 Detailed documentation: docs/MULTIMODAL.md

Reasoning Models

Universal support for reasoning models (DeepSeek R1, OpenAI o1, Volcengine Doubao-Seed-Code, etc.):

let request = ChatRequest {
    model: "deepseek-reasoner".to_string(),
    messages: vec![Message::text(Role::User, "Solve: 9.11 vs 9.9?")],
    ..Default::default()
};

let response = client.chat(&request).await?;
if let Some(reasoning) = response.reasoning_content {
    println!("Thinking: {}", reasoning);
}
println!("Answer: {}", response.content);

📖 Detailed documentation: docs/REASONING_MODELS_SUPPORT.md

Structured Output

use llm_connector::{ChatRequest, Message, ResponseFormat};

// JSON mode
let request = ChatRequest::new("gpt-4")
    .add_message(Message::user("List 3 languages as JSON array"))
    .with_response_format(ResponseFormat::json_object());

// JSON Schema mode (Structured Outputs)
let request = ChatRequest::new("gpt-4o")
    .add_message(Message::user("Extract person info"))
    .with_response_format(ResponseFormat::json_schema(
        "person",
        serde_json::json!({
            "type": "object",
            "properties": {
                "name": { "type": "string" },
                "age": { "type": "integer" }
            },
            "required": ["name", "age"]
        }),
    ));

Error Handling

use llm_connector::LlmConnectorError;

match client.chat(&request).await {
    Ok(response) => println!("Response: {}", response.content),
    Err(ref e) if e.is_rate_limited() => eprintln!("Rate limited, retry later"),
    Err(ref e) if e.should_reduce_context() => eprintln!("Input too long, reduce context"),
    Err(ref e) if e.is_auth_error() => eprintln!("Auth failed: {}", e),
    Err(ref e) if e.is_retryable() => eprintln!("Transient error, can retry: {}", e),
    Err(e) => eprintln!("Fatal error: {}", e),
}

Model Discovery

let client = LlmClient::openai("sk-...")?;
let models = client.models().await?;
println!("Available models: {:?}", models);

Examples

# Basic examples
cargo run --example openai_basic
cargo run --example anthropic_streaming
cargo run --example ollama_basic
cargo run --example xiaomi_basic

# Advanced examples
cargo run --example multimodal_basic
cargo run --example zhipu_tools
cargo run --example volcengine_streaming --features streaming

See examples/ directory for all available examples.

Documentation

Contributing

Contributions welcome! See docs/CONTRIBUTING.md for guidelines.

License

MIT