Supported Providers

ai-lib supports 20+ AI providers through a unified interface. Each provider is configured via environment variables and offers consistent API access. For enterprise environments requiring advanced provider management, routing, and analytics, see ai-lib-pro.

Provider Categories

Config-Driven Providers

These providers use a unified configuration system for easy setup and optimal performance:

Chinese Ecosystem Providers

Specialized for Chinese language and enterprise needs:

Independent Adapters

These providers have custom integration logic with high-performance direct HTTP client implementation:

Environment Variables

Configure API keys for each provider:

# Core providers
export OPENAI_API_KEY=your_openai_key
export GROQ_API_KEY=your_groq_key
export ANTHROPIC_API_KEY=your_anthropic_key
export GEMINI_API_KEY=your_gemini_key

# Chinese providers
export BAIDU_API_KEY=your_baidu_key
export TENCENT_SECRET_ID=your_tencent_id
export TENCENT_SECRET_KEY=your_tencent_key
export IFlytek_APP_ID=your_iflytek_id
export IFlytek_API_KEY=your_iflytek_key
export MOONSHOT_API_KEY=your_moonshot_key

# Other providers
export MISTRAL_API_KEY=your_mistral_key
export COHERE_API_KEY=your_cohere_key
export HUGGINGFACE_API_KEY=your_hf_key
export TOGETHERAI_API_KEY=your_together_key
export DEEPSEEK_API_KEY=your_deepseek_key
export QWEN_API_KEY=your_qwen_key
export XAI_API_KEY=your_xai_key
export OPENROUTER_API_KEY=your_openrouter_key
export REPLICATE_API_TOKEN=your_replicate_token
export PERPLEXITY_API_KEY=your_perplexity_key
export AI21_API_KEY=your_ai21_key
export ZHIPU_API_KEY=your_zhipu_key
export MINIMAX_API_KEY=your_minimax_key

# Ollama (local)
export OLLAMA_BASE_URL=http://localhost:11434

Provider Selection

Choose providers based on your needs:

use ai_lib::{AiClient, Provider};

// For speed and cost efficiency
let groq = AiClient::new(Provider::Groq)?;

// For reasoning and quality
let anthropic = AiClient::new(Provider::Anthropic)?;

// For multimodal capabilities
let gemini = AiClient::new(Provider::Gemini)?;

// For local/private deployment
let ollama = AiClient::new(Provider::Ollama)?;

Model Management

Each provider offers different models with varying capabilities:

// Get default model for provider
let default_model = client.default_chat_model();

// Use model arrays for load balancing (requires routing_mvp feature)
#[cfg(feature = "routing_mvp")]
{
    use ai_lib::{ModelArray, LoadBalancingStrategy};

    let array = ModelArray::new("production")
        .with_strategy(LoadBalancingStrategy::HealthBased)
        .add_endpoint(ModelEndpoint {
            name: "groq-1".into(),
            url: "https://api.groq.com".into(),
            weight: 1.0,
            healthy: true,
        });
}

Provider-Specific Notes

Groq

• Best for: Speed and cost efficiency
• Models: Llama 3, Mixtral variants
• Latency: Ultra-low (sub-100ms typical)
• Limitations: No multimodal support

Anthropic

• Best for: Reasoning and complex tasks
• Models: Claude 3 Opus, Sonnet, Haiku
• Strengths: Long context, tool use, safety
• Use cases: Research, analysis, complex reasoning

OpenAI

• Best for: General purpose, broad capabilities
• Models: GPT-4, GPT-3.5, DALL-E
• Strengths: Multimodal, function calling, reliability
• Use cases: General AI applications

Gemini

• Best for: Multimodal applications
• Models: Gemini Pro, Gemini Ultra
• Strengths: Native image/audio processing
• Use cases: Content analysis, multimodal AI

Enterprise Provider Management

For production environments with multiple providers, consider ai-lib-pro’s advanced features:

• Advanced Routing: Intelligent load balancing across providers
• Health Monitoring: Real-time provider health and performance tracking
• Cost Optimization: Provider cost analysis and automatic cost-based routing
• Failover Management: Automatic failover with configurable policies
• Usage Analytics: Detailed provider usage patterns and optimization insights

Provider Constants Reference

This section provides a complete reference of all supported providers, their enum constants, default models, and environment variable requirements. Use this as a quick reference when integrating ai-lib into your applications.

Provider Enum Constants

All providers are defined in the Provider enum:

use ai_lib::Provider;

// Config-driven providers
Provider::Groq
Provider::XaiGrok
Provider::Ollama
Provider::DeepSeek
Provider::Anthropic
Provider::AzureOpenAI
Provider::HuggingFace
Provider::TogetherAI
Provider::OpenRouter
Provider::Replicate

// Chinese ecosystem providers
Provider::BaiduWenxin
Provider::TencentHunyuan
Provider::IflytekSpark
Provider::Moonshot
Provider::ZhipuAI
Provider::MiniMax

// Independent adapter providers
Provider::OpenAI
Provider::Qwen
Provider::Gemini
Provider::Mistral
Provider::Cohere
Provider::Perplexity
Provider::AI21

Default Models

Each provider has a default chat model and optional multimodal model:

Environment Variables

Each provider requires specific environment variables for authentication:

Provider Capabilities Matrix

Feature Flags Reference

ai-lib uses Cargo feature flags to enable optional functionality. Use these flags in your Cargo.toml:

Core Features

[dependencies.ai-lib]
version = "0.4.0"
features = [
    # Enable interceptor pipeline (retry, rate limiters, circuit breaker)
    "interceptors",
    # Enable unified reqwest client factory/shared transport settings
    "unified_transport",
    # Enable unified SSE parser for streaming
    "unified_sse",
    # Enable minimal cost accounting metrics
    "cost_metrics",
    # Enable strategy-based routing MVP
    "routing_mvp",
    # Enable observability interfaces (Tracer, AuditSink)
    "observability",
    # Enable config hot-reload traits
    "config_hot_reload",
]

Convenience Aliases

# Enable streaming support (equivalent to "unified_sse")
features = ["streaming"]

# Enable transport layer (equivalent to "unified_transport")
features = ["transport"]

# Enable resilience features (equivalent to "interceptors")
features = ["resilience"]

# Enable hot reload (equivalent to "config_hot_reload")
features = ["hot_reload"]

# Enable all features
features = ["all"]

Recommended Feature Combinations

Feature Dependencies

• streaming → enables unified_sse
• transport → enables unified_transport
• resilience → enables interceptors
• hot_reload → enables config_hot_reload
• all → enables all core features

Next Steps

• Learn about Model Management for advanced provider strategies
• Explore Reliability Features for production deployments
• Check Advanced Examples for provider-specific patterns
• Discover ai-lib-pro for enterprise-grade provider management