xAI SDK

A Rust SDK for xAI's API, providing type-safe gRPC clients for all xAI services including Grok language models, embeddings, image generation, and more.

Features

Complete API Coverage: Full gRPC client implementation for all xAI services
Type Safety: Auto-generated Rust types from Protocol Buffers
Multiple Models: Support for all xAI language models (Grok-3, Grok-4, etc.)
Streaming Support: Real-time streaming for chat completions and text generation
Response Assembly: Convert streaming chunks into complete responses
Secure: TLS encryption with automatic certificate validation

Quick Start

Prerequisites

Rust 1.88+ installed
xAI API key

Installation

Add to your Cargo.toml:

[dependencies]
xai-sdk = "0.8"
tokio = { version = "1.0", features = ["full"] }
anyhow = "1.0"

Running the Examples

Set your API key as an environment variable:
```
export XAI_API_KEY="your-api-key-here"
```
Run the authentication info example:
```
cargo run --example auth_info
```
Run the raw text sampling example:
```
cargo run --example raw_text_sample
```

Run the chat completion example (supports multiple modes):

# Blocking completion
cargo run --example chat -- --complete

# Streaming completion
cargo run --example chat -- --stream

# Streaming with assembly
cargo run --example chat -- --assemble

Run the multi-client example (demonstrates using multiple services with shared channel):
```
cargo run --example multi_client
```

Run the interceptor composition example:

cargo run --example interceptor_compose

Run the tool calls example:
```
cargo run --example tool_calls
```

API Services

The SDK provides clients for all xAI services:

Chat Service

get_completion - Get chat completion
get_completion_chunk - Stream chat completion in chunks
start_deferred_completion - Start deferred chat completion
get_deferred_completion - Retrieve deferred completion
get_stored_completion - Get stored chat completion
delete_stored_completion - Delete stored chat completion

Sample Service

sample_text - Raw text generation
sample_text_streaming - Streaming text generation

Models Service

list_language_models - List available language models
list_embedding_models - List embedding models
list_image_generation_models - List image generation models

Embed Service

embed - Generate embeddings from text or images

Image Service

generate_image - Create images from text prompts

Video Service

generate_video - Create videos with deferred processing
get_deferred_video - Retrieve generated videos

Auth Service

get_api_key_info - Get API key information

Billing Service

set_billing_info - Set billing information for a team
get_billing_info - Get billing information for a team
list_payment_methods - List payment methods on file
set_default_payment_method - Set default payment method
get_amount_to_pay - Preview current billing period amount
analyze_billing_items - Analyze historical billing usage
list_invoices - List invoices for a team
list_prepaid_balance_changes - List prepaid credit balance changes
top_up_or_get_existing_pending_change - Top up prepaid credits
get_spending_limits - Get spending limits
set_soft_spending_limit - Set soft spending limit

Client Modules

The SDK is organized into focused modules, each providing easy client creation:

Available Modules

auth - Authentication services
billing - Billing and payment management
chat - Chat completions and streaming
documents - Document processing
embed - Text and image embeddings
image - Image generation
models - Model listing and information
sample - Text sampling and generation
tokenize - Text tokenization
video - Video generation with deferred processing

Complete Example

Here's a complete example showing multiple services using the modular architecture:

use anyhow::{Context, Result};
use std::env;
use xai_sdk::Request;
use xai_sdk::api::{
    Content, GetCompletionsRequest, GetModelRequest, Message, MessageRole, SampleTextRequest,
    content,
};
use xai_sdk::{chat, models, sample};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Load API key for authentication
    let api_key = env::var("XAI_API_KEY").context("XAI_API_KEY environment variable must be set")?;

    // Create authenticated clients for different services
    let mut models_client = models::client::new(api_key.clone()).await?;
    let mut sample_client = sample::client::new(api_key.clone()).await?;
    let mut chat_client = chat::client::new(api_key).await?;

    // List available models
    let models_request = Request::new(());
    let models_response = models_client.list_language_models(models_request).await?;
    println!("Available models: {:?}", models_response.into_inner().models);

    // Get details for a specific model
    let model_request = Request::new(GetModelRequest {
        name: "grok-3-latest".to_string(),
    });
    let model_response = models_client.get_language_model(model_request).await?;
    println!("Model details: {:?}", model_response.into_inner());

    // Generate text
    let sample_request = Request::new(SampleTextRequest {
        prompt: vec!["Hello, world!".to_string()],
        model: "grok-3-latest".to_string(),
        ..Default::default()
    });
    let sample_response = sample_client.sample_text(sample_request).await?;
    println!("Generated: {}", sample_response.into_inner().choices[0].text);

    // Chat completion
    let message = Message {
        role: MessageRole::RoleUser.into(),
        content: vec![Content {
            content: Some(content::Content::Text("Explain Rust ownership".to_string())),
        }],
        ..Default::default()
    };
    let chat_request = Request::new(GetCompletionsRequest {
        model: "grok-3-latest".to_string(),
        messages: vec![message],
        ..Default::default()
    });
    let chat_response = chat_client.get_completion(chat_request).await?;
    println!(
        "Chat response: {}",
        chat_response.into_inner().outputs[0]
            .message
            .as_ref()
            .unwrap()
            .content
    );

    Ok(())
}

Streaming Utilities

The SDK provides powerful utilities for working with streaming responses:

Stream Consumer

A flexible callback system for processing streaming data:

on_chunk(chunk) - Called for each complete chunk received
on_reasoning_start(&OutputContext) - Called once per output when the reasoning phase starts (before the first reasoning token)
on_reasoning_token(&OutputContext, token: &str) - Called for each piece of reasoning content
on_reasoning_complete(&OutputContext) - Called once when the reasoning phase completes for an output
on_content_start(&OutputContext) - Called once per output when the content phase starts (before the first content token)
on_content_token(&OutputContext, token: &str) - Called for each piece of response content
on_content_complete(&OutputContext) - Called once when the content phase completes for an output
on_inline_citations(&OutputContext, &[InlineCitation]) - Called when inline citations are present in a delta
on_client_tool_calls(&OutputContext, &[ToolCall]) - Called when client-side tool calls are present
on_server_tool_calls(&OutputContext, &[ToolCall]) - Called when server-side tool calls are present
on_usage(&SamplingUsage) - Called once on the last chunk with usage statistics
on_citations(&[String]) - Called once on the last chunk with citations

The OutputContext provides:

total_outputs - Total number of outputs in the stream
output_index - Index of the output this context belongs to
reasoning_status - Current status of the reasoning phase (Init, Start, Pending, or Complete)
content_status - Current status of the content phase (Init, Start, Pending, or Complete)

Stream Processing Functions

chat::stream::process - Process streaming responses with custom callbacks
chat::stream::assemble - Convert collected chunks into complete responses
chat::stream::Consumer::new_static() - Empty consumer with 'static lifetime for one-line chaining: Consumer::new_static().on_content_token(...).on_usage(...)
chat::stream::Consumer::with_sink(snk) - Consumer that forwards all stream activity as Events into any Sink<Event> (e.g. futures::channel::mpsc::unbounded()). Create the channel, pass the sender, then drain the receiver for a single event stream. Bounded senders apply backpressure.
chat::stream::Event - Enum of streaming events: Chunk, ReasoningStart/ReasoningToken/ReasoningComplete, ContentStart/ContentToken/ContentComplete, InlineCitations, ClientToolCalls, ServerToolCalls, Citations, Usage, Complete, Error.
chat::stream::Consumer::with_stdout() - Pre-configured consumer for single-output real-time output
chat::stream::Consumer::with_buffered_stdout() - Pre-configured consumer for multi-output buffered output

Interceptors

The SDK provides a flexible interceptor system for customizing request handling:

Authentication Interceptor

The auth() function creates an interceptor that adds Bearer token authentication:

use xai_sdk::common::interceptor::auth;

let interceptor = auth("your-api-key");
let client = chat::client::with_interceptor(interceptor).await?;

Composing Interceptors

Combine multiple interceptors using compose():

use xai_sdk::common::interceptor::{auth, compose};

let interceptors: Vec<Box<dyn xai_sdk::export::service::Interceptor + Send + Sync>> = vec![
    Box::new(auth("your-api-key")),
    Box::new(|mut req| {
        req.metadata_mut().insert("x-custom-header", "value".parse().unwrap());
        Ok(req)
    }),
];
let composed = compose(interceptors);
let client = chat::client::with_interceptor(composed).await?;

Note: All interceptors must be Send + Sync to ensure thread safety when used in async contexts.

ClientInterceptor Type

All client functions return ClientInterceptor, a concrete type that can be:

Stored in structs
Used in trait implementations
Passed between functions without type erasure issues
Used across thread boundaries (Send + Sync) - safe to use in tokio::spawn and other async contexts

The ClientInterceptor can be created from any impl Interceptor + Send + Sync + 'static:

use xai_sdk::common::interceptor::ClientInterceptor;

let interceptor = ClientInterceptor::new(|mut req| {
    // Custom logic
    Ok(req)
});

// Can be used in spawned tasks
tokio::spawn(async move {
    let client = chat::client::with_interceptor(interceptor).await?;
    // Use client...
});

Configuration

The SDK supports comprehensive configuration options:

Temperature: Controls randomness (0.0 to 2.0)
Top-p: Nucleus sampling parameter (0.0 to 1.0)
Max tokens: Maximum tokens to generate
Log probabilities: Enable detailed token probability logging
Multiple completions: Generate multiple responses per request
Stop sequences: Custom stop conditions
Frequency/Presence penalties: Control repetition and topic diversity

Security

TLS Encryption: Automatic HTTPS with certificate validation
Authentication: Bearer token support for API key authentication
Secure by Default: No manual TLS configuration required

Error Handling

Comprehensive error handling for:

Connection errors and timeouts
Authentication failures
API rate limiting
Invalid parameters
Network issues

Development

This SDK is built using:

Protocol Buffers: Auto-generated Rust types from xAI's .proto definitions
Tonic: Modern gRPC framework for Rust with async/await support
Prost: High-performance Protocol Buffer implementation

The code is generated from xAI's official Protocol Buffer definitions, ensuring compatibility and type safety.

Changelog

See CHANGELOG.md for a detailed list of changes and new features.

License

This project is licensed under the MIT License.

xai-sdk 0.8.8