⚠️ UNSTABLE — This crate is in active development. APIs may change without notice.

Platform-specific local AI runtimes for the Rusty AI SDK.

Each runtime is gated behind a Cargo feature flag:

• browser — WASM browser AI (Chrome/Edge built-in AI)
• gemini-nano — Android Gemini Nano (Prompt API via JNI)
• foundationmodels — Apple Foundation Models (uniffi Swift bindings)
• phi-silica — Windows Phi Silica (C# bridge via build.rs)

rs_ai — Rust AI SDK

A comprehensive Rust SDK for building AI applications with cloud and local providers, streaming, 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(())
}

Providers

Local Runtimes

Add rs_ai_local to use on-device AI:

[dependencies]
rs_ai_local = { version = "0.2", features = ["gemini-nano"] }

Browser (Chrome Prompt API)

Enable the browser feature and compile for wasm32 to use the browser's built-in AI APIs:

rs_ai_local = { version = "0.2", features = ["browser"] }

Use from Rust WASM:

use rs_ai_local::browser::{wasm_bridge::WasmBrowserBridge, BrowserAiOptions};

let bridge = WasmBrowserBridge;
let answer = bridge
    .generate("Explain Rust ownership in one paragraph", &BrowserAiOptions::default())
    .await?;

Build for the web:

wasm-pack build --target web --features browser

Android (Gemini Nano)

Enable the feature and add initialization to your Activity. They call our init function — just one line of Kotlin:

// MainActivity.kt
class MainActivity : Activity() {
    companion object {
        init { System.loadLibrary("rs_ai_local") }
        @JvmStatic external fun init(context: Context)
    }
    
    override fun onCreate(savedInstanceState: Bundle?) {
        super.onCreate(savedInstanceState)
        init(this)
    }
}

Then use from Rust — your code:

use rs_ai_local::gemini_nano::init_with_context;
use rs_ai_local::gemini_nano::GeminiNanoProvider;

// In your Rust code
let provider = GeminiNanoProvider::new(my_bridge);
let response = provider.model().generate("Hello!").await?;

macOS (Foundation Models)

Enable the feature. Build on macOS with Xcode — build.rs auto-compiles the Swift bridge:

rs_ai_local = { version = "0.2", features = ["foundationmodels"] }

Use from Rust:

use rs_ai_local::foundationmodels::{is_available, respond};

if is_available() {
    let answer = respond("What is Rust?").await?;
}

Windows (Phi Silica)

Enable the feature. Build with .NET SDK — build.rs auto-compiles the C# bridge:

rs_ai_local = { version = "0.2", features = ["phi-silica"] }

Use from Rust:

use rs_ai_local::phi_silica::respond;

let answer = respond("Hello!").await?;

Examples

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?);
}

Testing

cargo test

License

MPL-2.0