Crate ai_types

ai-types

Write AI applications that work with any provider 🚀

This crate provides unified trait abstractions for AI models, letting you write code once and switch between providers (OpenAI, Anthropic, local models, etc.) without changing your application logic.

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   Your App      │───▶│    ai-types      │◀───│   Providers     │
│                 │    │   (this crate)   │    │                 │
│ - Chat bots     │    │                  │    │ - openai        │
│ - Search        │    │ - LanguageModel  │    │ - anthropic     │
│ - Content gen   │    │ - EmbeddingModel │    │ - llama.cpp     │
│ - Voice apps    │    │ - ImageGenerator │    │ - whisper       │
└─────────────────┘    └──────────────────┘    └─────────────────┘

Supported AI Capabilities

Capability	Trait	Description
Language Models	LanguageModel	Text generation, conversations, structured output
Text Streaming	TextStream	Unified interface for streaming text responses
Embeddings	EmbeddingModel	Convert text to vectors for semantic search
Image Generation	ImageGenerator	Create images with progressive quality improvement
Text-to-Speech	AudioGenerator	Generate speech audio from text
Speech-to-Text	AudioTranscriber	Transcribe audio to text
Content Moderation	Moderation	Detect policy violations with confidence scores

Examples

Basic Chat Bot

use ai_types::{LanguageModel, llm::{Message, Request}};
use futures_lite::StreamExt;

async fn chat_example(model: impl LanguageModel) -> ai_types::Result {
    let messages = [
        Message::system("You are a helpful assistant"),
        Message::user("What's the capital of France?")
    ];
     
    let request = Request::new(messages);
    let mut response = model.respond(request);
     
    Ok(response.await?)
}

Structured Output with Tools

use ai_types::{LanguageModel, llm::{Message, Request, Tool}};
use serde::{Deserialize, Serialize};
use schemars::JsonSchema;

#[derive(JsonSchema, Deserialize, Serialize)]
struct WeatherQuery {
    location: String,
    units: Option<String>,
}

struct WeatherTool;

impl Tool for WeatherTool {
    const NAME: &str = "get_weather";
    const DESCRIPTION: &str = "Get current weather for a location";
    type Arguments = WeatherQuery;
     
    async fn call(&mut self, args: Self::Arguments) -> ai_types::Result {
        Ok(format!("Weather in {}: 22°C, sunny", args.location))
    }
}

async fn weather_bot(model: impl LanguageModel) -> ai_types::Result {
    let request = Request::new(vec![
        Message::user("What's the weather like in Tokyo?")
    ]).with_tool(WeatherTool);
     
    // Model can now call the weather tool automatically
    let response: String = model.generate(request).await?;
    Ok(response)
}

See llm::tool for more details on using tools with language models.

Semantic Search with Embeddings

use ai_types::EmbeddingModel;

async fn find_similar_docs(
    model: impl EmbeddingModel,
    query: &str,
    documents: &[&str]
) -> ai_types::Result<Vec<f32>> {
    // Convert query to vector
    let query_embedding = model.embed(query).await?;
     
    // In a real app, you'd compare with document embeddings
    // and find the most similar ones using cosine similarity
     
    Ok(query_embedding)
}

Progressive Image Generation

use ai_types::{ImageGenerator, image::{Prompt, Size}};
use futures_lite::StreamExt;

async fn generate_image(generator: impl ImageGenerator) -> Result<Vec<u8>, Box<dyn std::error::Error>> {
    let prompt = Prompt::new("A beautiful sunset over mountains");
    let size = Size::square(1024);
     
    let mut image_stream = generator.create(prompt, size);
    let mut final_image = Vec::new();
     
    // Each iteration gives us a complete image with progressively better quality
    while let Some(image_result) = image_stream.next().await {
        let current_image = image_result?;
        final_image = current_image; // Keep the latest (highest quality) version
         
        // Optional: Display preview of current quality level
        println!("Received image update, {} bytes", final_image.len());
    }
     
    Ok(final_image) // Return the final highest-quality image
}

Modules

audio

Audio generation and transcription.

embedding

Text embeddings.

image

Text-to-image generation.

llm

Language Models and Conversation Management

moderation

Content moderation utilities.

Structs

Error

The Error type, a wrapper around a dynamic error type.

Traits

AudioGenerator

Generates audio from text prompts.

AudioTranscriber

Transcribes audio to text.

EmbeddingModel

Converts text to vector representations.

ImageGenerator

Trait for generating and editing images from prompts and masks.

LanguageModel

Language models for text generation and conversation.

Moderation

Trait for content moderation services.

TextStream

A trait for streaming text responses from language models.

Type Aliases

Result

Result type used throughout the crate.

Attribute Macros

tool

Converts an async function into an AI tool that can be called by language models.