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:

Provider	Chat	Streaming	Multimodal	Function Calls	Notes
Groq	✅	✅	❌	✅	Ultra-low latency, Llama models
Anthropic	✅	✅	✅	✅	Claude 3, strong reasoning
Azure OpenAI	✅	✅	✅	✅	Enterprise OpenAI with compliance
HuggingFace	✅	✅	Varies	Varies	Community models, varies by model
TogetherAI	✅	✅	Varies	Varies	Cost-efficient, open models
DeepSeek	✅	✅	❌	✅	Reasoning-focused models
Ollama	✅	✅	Varies	Varies	Local/airgapped deployment
xAI Grok	✅	✅	❌	✅	Real-time oriented
OpenRouter	✅	✅	Varies	Varies	OpenAI-compatible gateway, provider/model routing
Replicate	✅	✅	Varies	Varies	Hosted open-source models

Chinese Ecosystem Providers

Specialized for Chinese language and enterprise needs:

Provider	Chat	Streaming	Multimodal	Function Calls	Notes
Baidu Wenxin	✅	✅	✅	✅	Enterprise Chinese AI
Tencent Hunyuan	✅	✅	✅	✅	Cloud integration
iFlytek Spark	✅	✅	✅	✅	Voice + multimodal
Moonshot Kimi	✅	✅	✅	✅	Long context models

Independent Adapters

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

Provider	Chat	Streaming	Multimodal	Function Calls	Notes
OpenAI	✅	✅	✅	✅	GPT-4, GPT-3.5, broad model set
Google Gemini	✅	✅	✅	✅	Native multimodal, Gemini Pro
Mistral	✅	✅	Partial	✅	European models, lightweight
Cohere	✅	✅	Limited	✅	Command models, RAG optimized
Qwen	✅	✅	✅	✅	Alibaba’s multilingual models
Perplexity	✅	✅	Limited	✅	Search-augmented conversations
AI21	✅	✅	Limited	✅	Jurassic series
ZhipuAI (GLM)	✅	✅	✅	✅	Chinese GLM series
MiniMax	✅	✅	✅	✅	Chinese multimodal

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();

// Check model capabilities
let capabilities = client.model_capabilities(&model_name);

// Use model arrays for load balancing
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

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