Crate ai_sdk_provider 

AI SDK Provider Specification (v3)

Provider interface specification defining trait-based contracts for AI model implementations. This crate establishes the core abstractions that enable provider-agnostic AI applications through standardized interfaces for language models, embeddings, image generation, speech synthesis, transcription, and document reranking.

Architecture

The specification employs a trait-based design where each AI capability is represented by a dedicated trait (e.g., LanguageModel, EmbeddingModel). Provider implementations satisfy these traits to expose their services through a uniform API surface.

Supported Model Types

• Language Models - Text generation, chat completion, and conversational AI
• Embedding Models - Vector representations for semantic text similarity
• Image Models - Text-to-image and image-to-image generation
• Speech Models - Text-to-speech synthesis with voice customization
• Transcription Models - Speech-to-text conversion with timestamps
• Reranking Models - Semantic document reordering by relevance

Usage

Provider traits expose low-level do_* methods for direct model interaction and high-level builder methods for ergonomic usage with optional parameters.

use ai_sdk_provider::{LanguageModel, CallOptions, Message, UserContentPart};

async fn generate_text<M: LanguageModel>(model: &M) {
    let options = CallOptions {
        prompt: vec![Message::User {
            content: vec![UserContentPart::Text {
                text: "Explain quantum entanglement".to_string(),
            }],
        }].into(),
        max_output_tokens: Some(500),
        temperature: Some(0.7),
        ..Default::default()
    };

    let response = model.do_generate(options).await.unwrap();
    println!("Generated: {:?}", response.content);
}

Provider Implementations

Reference implementations maintained in companion crates:

• ai-sdk-openai - OpenAI GPT, DALL-E, Whisper, and embeddings

Version Compatibility

This crate implements specification version 3. Provider implementations must return “v3” from their specification_version() method to ensure compatibility with this interface.

Re-exports

pub use embedding_model::EmbedOptions;

pub use embedding_model::EmbedResponse;

pub use embedding_model::Embedding;

pub use embedding_model::EmbeddingModel;

pub use embedding_model::EmbeddingUsage;

pub use image_model::CallWarning as ImageCallWarning;

pub use image_model::ImageData;

pub use image_model::ImageGenerateOptions;

pub use image_model::ImageGenerateResponse;

pub use image_model::ImageModel;

pub use image_model::ImageProviderMetadata;

pub use json_value::JsonArray;

pub use json_value::JsonObject;

pub use json_value::JsonValue;

pub use language_model::CallOptions;

pub use language_model::Content;

pub use language_model::FinishReason;

pub use language_model::GenerateResponse;

pub use language_model::LanguageModel;

pub use language_model::StreamPart;

pub use language_model::StreamResponse;

pub use language_model::Usage;

pub use provider::ProviderV3;

pub use reranking_model::Documents;

pub use reranking_model::RankingItem;

pub use reranking_model::RerankOptions;

pub use reranking_model::RerankResponse;

pub use reranking_model::RerankingModel;

pub use reranking_model::ResponseInfo as RerankingResponseInfo;

pub use shared::SharedHeaders;

pub use shared::SharedProviderMetadata;

pub use shared::SharedProviderOptions;

pub use shared::SharedWarning;

pub use speech_model::AudioData;

pub use speech_model::CallWarning as SpeechCallWarning;

pub use speech_model::SpeechGenerateOptions;

pub use speech_model::SpeechGenerateResponse;

pub use speech_model::SpeechModel;

pub use transcription_model::AudioInput;

pub use transcription_model::CallWarning as TranscriptionCallWarning;

pub use transcription_model::RequestInfo as TranscriptionRequestInfo;

pub use transcription_model::ResponseInfo as TranscriptionResponseInfo;

pub use transcription_model::TranscriptionModel;

pub use transcription_model::TranscriptionOptions;

pub use transcription_model::TranscriptionResponse;

pub use transcription_model::TranscriptionSegment;

pub use error::Error;

pub use error::Result;

Modules

embedding_model

Embedding model interfaces and types for text embedding generation. Embedding Model v3 Specification Implementation

error

Error types used across various operations.

image_model

Image model interfaces and types for image generation. Image generation model provider interface and types.

json_value

JSON value types for provider metadata and structured data.

language_model

Language model interfaces and types for text generation and chat completion. Language Model v3 specification types

provider

Provider trait for AI model factories. Provider trait for AI model factories.

reranking_model

Reranking model interfaces and types for document reranking. Reranking model provider interface and types.

shared

Shared types and utilities used across all model types.

speech_model

Speech model interfaces and types for text-to-speech synthesis. Speech synthesis model provider interface and types.

transcription_model

Transcription model interfaces and types for speech-to-text transcription. Audio transcription model provider interface and types.