LLM Kit Baseten

Baseten provider for LLM Kit - Complete integration with Baseten's Model APIs and custom model deployments.

Note: This provider uses the standardized builder pattern. See the Quick Start section for the recommended usage.

Features

• Text Generation: Generate text using Baseten-hosted models or custom deployments
• Streaming: Stream responses in real-time with support for tool calls
• Tool Calling: Support for function calling with both hosted and custom models
• Text Embedding: Generate embeddings using custom model deployments
• Model APIs: Access to hosted models like DeepSeek V3, Kimi K2, and Qwen 3
• Custom Deployments: Support for dedicated model deployments with custom URLs

Installation

Add this to your Cargo.toml:

[dependencies]
llm-kit-baseten = "0.1"
llm-kit-core = "0.1"
llm-kit-provider = "0.1"
tokio = { version = "1", features = ["full"] }

Quick Start

Using the Client Builder (Recommended)

use llm_kit_baseten::BasetenClient;
use llm_kit_provider::language_model::LanguageModel;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create provider using the client builder
    let provider = BasetenClient::new()
        .api_key("your-api-key")  // Or use BASETEN_API_KEY env var
        .build();
    
    // Create a language model
    let model = provider.chat_model(Some("deepseek-ai/DeepSeek-V3-0324"));
    
    println!("Model: {}", model.model_id());
    println!("Provider: {}", model.provider());
    Ok(())
}

Using Settings Directly (Alternative)

use llm_kit_baseten::{BasetenProvider, BasetenProviderSettings};
use llm_kit_provider::language_model::LanguageModel;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Create provider with settings
    let provider = BasetenProvider::new(BasetenProviderSettings::default());
    
    let model = provider.chat_model(Some("deepseek-ai/DeepSeek-V3-0324"));
    
    println!("Model: {}", model.model_id());
    Ok(())
}

Configuration

Environment Variables

Set your Baseten API key as an environment variable:

export BASETEN_API_KEY=your-api-key

Using the Client Builder

use llm_kit_baseten::BasetenClient;

let provider = BasetenClient::new()
    .api_key("your-api-key")
    .base_url("https://inference.baseten.co/v1")
    .header("Custom-Header", "value")
    .build();

Builder Methods

The BasetenClient builder supports:

• .api_key(key) - Set the API key (overrides BASETEN_API_KEY environment variable)
• .base_url(url) - Set the base URL for Model APIs (default: https://inference.baseten.co/v1)
• .model_url(url) - Set a custom model URL for dedicated deployments
• .name(name) - Set provider name (optional)
• .header(key, value) - Add a single custom header
• .headers(map) - Add multiple custom headers
• .build() - Build the provider

Provider-Specific Options

Model APIs vs Custom Deployments

Baseten supports two deployment modes:

Model APIs (Hosted Models)

Use the default base URL to access Baseten's hosted models:

use llm_kit_baseten::BasetenClient;

let provider = BasetenClient::new()
    .api_key("your-api-key")
    .build();

// Specify model ID from hosted models
let model = provider.chat_model(Some("deepseek-ai/DeepSeek-V3-0324"));

Custom Model Deployments

For dedicated deployments, specify a custom model URL:

use llm_kit_baseten::BasetenClient;

let provider = BasetenClient::new()
    .api_key("your-api-key")
    .model_url("https://model-{id}.api.baseten.co/environments/production/sync/v1")
    .build();

// Model ID is optional when using custom URL
let model = provider.chat_model(None);

Important:

• Chat models require /sync/v1 endpoints
• Embedding models require /sync or /sync/v1 endpoints

Text Embeddings

Embeddings require a custom model URL and are not available via Model APIs:

use llm_kit_baseten::BasetenClient;

let provider = BasetenClient::new()
    .api_key("your-api-key")
    .model_url("https://model-{id}.api.baseten.co/environments/production/sync")
    .build();

let model = provider.text_embedding_model(None);

Supported Models

Model APIs (Hosted Models)

All Baseten Model API models are supported, including:

• DeepSeek: deepseek-ai/DeepSeek-R1-0528, deepseek-ai/DeepSeek-V3-0324, deepseek-ai/DeepSeek-V3.1
• Kimi: moonshotai/Kimi-K2-Instruct-0905
• Qwen: Qwen/Qwen3-235B-A22B-Instruct-2507, Qwen/Qwen3-Coder-480B-A35B-Instruct
• OpenAI: openai/gpt-oss-120b
• GLM: zai-org/GLM-4.6

Custom Models

Any model deployed on Baseten can be used with a custom model URL. Supported model types:

• Chat Models - Text generation with optional tool calling
• Embedding Models - Text embeddings for similarity search

For a complete list of available hosted models, see the Baseten Model APIs documentation.

Examples

See the examples/ directory for complete examples:

• chat.rs - Basic chat completion with Model APIs
• stream.rs - Streaming responses
• chat_tool_calling.rs - Tool calling with function definitions
• stream_tool_calling.rs - Streaming with tool calls
• text_embedding.rs - Text embeddings with custom deployments

Run examples with:

export BASETEN_API_KEY=your-api-key
cargo run --example chat
cargo run --example stream
cargo run --example chat_tool_calling
cargo run --example stream_tool_calling
cargo run --example text_embedding

Documentation

• API Documentation
• LLM Kit Documentation
• Baseten API Reference
• Baseten Model APIs

License

Apache-2.0

Contributing

Contributions are welcome! Please see the Contributing Guide for more details.