Crate ai_lib_rust 

ai-lib-rust

这是 AI-Protocol 规范的高性能 Rust 参考实现，提供统一的多厂商 AI 模型交互接口。

Protocol Runtime for AI-Protocol - A high-performance Rust reference implementation that enables provider-agnostic AI model interactions.

Overview

This library implements the AI-Protocol specification as a runtime, where all logic is operators and all configuration is protocol. It provides a unified interface for interacting with AI models across different providers without hardcoding provider-specific logic.

Core Philosophy

• Protocol-Driven: All behavior is configured through protocol manifests, not code
• Provider-Agnostic: Unified interface across OpenAI, Anthropic, Google, and others
• Streaming-First: Native support for Server-Sent Events (SSE) streaming
• Type-Safe: Strongly typed request/response handling with comprehensive error types

Key Features

• Unified Client: AiClient provides a single entry point for all AI interactions
• Protocol Loading: Load and validate protocol manifests from local files or remote URLs
• Streaming Pipeline: Configurable operator pipeline for response processing
• Batching: Efficient request batching with [batch::BatchCollector] (requires batch feature)
• Caching: Response caching with pluggable backends via cache module
• Resilience: Circuit breaker and rate limiting via resilience module
• Content Safety: Guardrails for content filtering via [guardrails] module
• Telemetry: Optional feedback collection via [telemetry] module

Quick Start

use ai_lib_rust::{AiClient, AiClientBuilder, Message, MessageRole};

#[tokio::main]
async fn main() -> ai_lib_rust::Result<()> {
    let client = AiClientBuilder::new()
        .with_protocol_path("protocols/openai.yaml")?
        .with_api_key("your-api-key")
        .build()?;

    let messages = vec![
        Message::user("Hello, how are you?"),
    ];

    // Streaming response
    let mut stream = client.chat_stream(&messages, None).await?;
    // Process stream events...
     
    Ok(())
}

Module Organization

Module	Description
protocol	Protocol specification loading and validation
client	AI client implementation and builders
pipeline	Streaming response pipeline operators
types	Core type definitions (messages, events, tools)
[batch]	Request batching and parallel execution (requires batch feature)
cache	Response caching with multiple backends
[embeddings]	Embedding generation and vector operations (requires embeddings feature)
resilience	Circuit breaker and rate limiting
[guardrails]	Content filtering and safety checks (requires guardrails feature)
[tokens]	Token counting and cost estimation (requires tokens feature)
[telemetry]	Optional feedback and telemetry collection (requires telemetry feature)

Re-exports

pub use client::CallStats;

pub use client::CancelHandle;

pub use client::ClientMetrics;

pub use client::ChatBatchRequest;

pub use client::EndpointExt;

pub use client::AiClient;

pub use client::AiClientBuilder;

pub use feedback::FeedbackEvent;

pub use feedback::FeedbackSink;

pub use types::events::StreamingEvent;

pub use types::message::Message;

pub use types::message::MessageRole;

pub use types::tool::ToolCall;

pub use error::Error;

pub use error::ErrorContext;

pub use error_code::StandardErrorCode;

Modules

cache

响应缓存模块：提供可插拔的缓存后端以减少重复 API 调用。

client

统一客户端接口：提供协议驱动的 AI 模型交互入口。

drivers

Provider 驱动抽象层 — 通过 trait 实现多厂商 API 适配的动态分发

error

Error type for the library 错误处理模块：提供统一的错误类型和结构化错误上下文。

error_code

V2 标准错误码：定义 13 个规范错误码及其重试/回退语义。

feedback

核心反馈类型：提供 FeedbackSink trait 和多种反馈事件（始终编译）。

pipeline

流水线处理模块：实现流式响应处理的核心算子执行引擎。

plugins

Plugin and middleware system.

protocol

协议规范层：负责加载、验证和管理 AI-Protocol 规范文件。

registry

能力注册表 — 根据 Manifest 声明动态加载和管理运行时模块

resilience

弹性模式模块：提供熔断器和限流器等可靠性保障机制。

structured

Structured output module for ai-lib-rust.

transport

types

类型系统模块：定义基于 AI-Protocol 规范的核心数据类型。

utils

Utility modules

Type Aliases

BoxStream

A unified pinned, boxed stream that emits PipeResult<T>

PipeResult

A specialized Result for pipeline operations

Result

Result type alias for the library