rs_ai — Rust AI SDK (v0.2.9)

Comprehensive Rust SDK for AI applications. Cloud + local providers, streaming, agent loop, image/video generation, realtime voice, and a clean async-first API.

Quick Start

[dependencies]
rs_ai = "0.2"

use rs_ai::rs_ai_claude;

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let answer = rs_ai_claude()
        .model("claude-sonnet-4-6")
        .generate("What is 2+2?")
        .await?;
    println!("{}", answer);
    Ok(())
}

Streaming

use rs_ai::rs_ai_claude;
use futures::StreamExt;

let mut stream = rs_ai_claude().stream("Write a poem").await?;
while let Some(chunk) = stream.next().await {
    print!("{}", chunk?);
}

Agent Loop (auto tool execution)

use rs_ai::rs_ai_chatgpt;
use rs_ai_core::agent_loop;
use rs_ai_core::tool::{ToolSet, Tool};

let mut tools = ToolSet::new();
tools.add(MyWeatherTool);

let result = agent_loop(&model, "What's the weather?".into(),
    Default::default().max_steps(5), Some(&tools)).await?;
// result.steps holds each intermediate step

Provider Registry

use rs_ai_core::create_provider_registry;
use rs_ai_providers::chatgpt::ChatGptProvider;

let registry = create_provider_registry()
    .register("openai", ChatGptProvider::new("sk-..."));
let model = registry.model("openai/gpt-4o")?;

Image Generation

let image = rs_ai_chatgpt()
    .model("dall-e-3")
    .generate_image("A cat in space", Default::default())
    .await?;
// image.data holds the bytes

Providers

Function	Provider	Env Var	Extras
rs_ai_claude()	Anthropic Claude	ANTHROPIC_API_KEY	streaming, tools, vision
rs_ai_chatgpt()	OpenAI ChatGPT	OPENAI_API_KEY	DALL-E, Sora, TTS, STT, realtime
rs_ai_gemini()	Google Gemini	GOOGLE_API_KEY	Imagen, Veo, live API
rs_ai_xai()	xAI Grok	XAI_API_KEY	Grok Imagine
rs_ai_cloudflare(id)	Cloudflare Workers AI	CLOUDFLARE_API_TOKEN	—
rs_ai_compatible(url)	Any OpenAI-compatible	OPENAI_API_KEY	OpenRouter, vLLM, Ollama, etc.

Real TTS/STT

let audio = rs_ai::chatgpt()
    .model("tts-1")
    .speak("Hello world!")  // Returns real MP3 bytes via OpenAI TTS
    .await?;

let text = rs_ai::chatgpt()
    .model("whisper-1")
    .transcribe(audio_bytes, "audio/webm")
    .await?;

Reranking (Cohere / Voyage)

use rs_ai_providers::cohere::CohereRerankingModel;
use rs_ai_core::RerankOptions;

let model = CohereRerankingModel::new("rerank-v3.5", api_key);
let result = model.rerank("my query", vec!["doc1".into(), "doc2".into()], RerankOptions::default()).await?;

Local Runtimes

[dependencies]
rs_ai_local = { version = "0.2", features = ["browser"] }       # WASM browser AI
rs_ai_local = { version = "0.2", features = ["gemini-nano"] }   # Android Gemini Nano
rs_ai_local = { version = "0.2", features = ["foundationmodels"] } # macOS Apple FM
rs_ai_local = { version = "0.2", features = ["phi-silica"] }    # Windows Phi Silica

Testing

make ci        # fmt + check + clippy + test + doc (152 tests)
cargo test     # same

License

MPL-2.0