​
What are providers?

Providers in Agenite are implementations of the LLMProvider interface that connect your agents to specific large language model services. They translate the standardized Agenite messaging format into specific API calls and handle the response processing back into the Agenite format. This provider abstraction allows your agents to:
  • Switch between different LLM services without changing your core agent code
  • Take advantage of specialized capabilities of different providers
  • Maintain a consistent development experience across different LLMs
Key aspects of providers
  • Implementation of the LLM interface: Each provider implements the standard LLMProvider interface
  • Provider-specific configuration: Handle authentication, model selection, and service-specific options
  • Consistent usage pattern: All providers are used in the same way within agent definitions
  • Easy switching: Agents can be reconfigured to use different providers with minimal code changes
  • Specialized optimizations: Providers can implement optimizations for their specific LLM service

​
Supported providers

Agenite includes official support for several popular LLM providers:

​
OpenAI

The OpenAI provider connects to OpenAI’s GPT models, including GPT-3.5 Turbo and GPT-4. 
import { OpenAI } from '@agenite/openai';

const provider = new OpenAI({
  apiKey: 'your-api-key',
  model: 'gpt-4-turbo',
  temperature: 0.7
});

​
Anthropic

The Anthropic provider connects to Anthropic’s Claude models. 
import { Anthropic } from '@agenite/anthropic';

const provider = new Anthropic({
  apiKey: 'your-api-key',
  model: 'claude-3-sonnet-20240229',
  temperature: 0.5
});

​
AWS Bedrock

The AWS Bedrock provider connects to various models available on the AWS Bedrock service, including Claude, Llama, and more. 
import { BedrockProvider } from '@agenite/bedrock';

const provider = new BedrockProvider({
  model: 'anthropic.claude-3-5-sonnet-20240620-v1:0',
  temperature: 0.7
});

​
Ollama

The Ollama provider connects to locally-hosted open-source models using Ollama. 
import { Ollama } from '@agenite/ollama';

const provider = new Ollama({
  model: 'llama3.2',
  baseUrl: 'http://localhost:11434',
  temperature: 0.7
});

​
Provider configuration

Each provider has its own configuration options, but they typically include:
  1. Authentication credentials: API keys or other authentication methods
  2. Model selection: Which specific model to use
  3. Generation parameters: Temperature, max tokens, etc.
  4. Endpoint configuration: Custom endpoints or regions
Here’s an example of the configuration options for the OpenAI provider: 
interface OpenAIConfig {
  apiKey: string;
  model: string;
  temperature?: number;
  maxTokens?: number;
  topP?: number;
  frequencyPenalty?: number;
  presencePenalty?: number;
  stop?: string[];
  organization?: string;
}

​
Using providers with agents

Providers are passed to agents during initialization: 
import { Agent } from '@agenite/agent';
import { OpenAI } from '@agenite/openai';
import { Calculator } from '@agenite/tool';

const agent = new Agent({
  name: 'math-assistant',
  provider: new OpenAI({ 
    apiKey: 'your-api-key', 
    model: 'gpt-4-turbo' 
  }),
  tools: [new Calculator()],
  instructions: 'You are a helpful math assistant.',
});

const result = await agent.execute({
  messages: [{ role: 'user', content: 'What is 137 * 456?' }]
});
The agent uses the provider to handle all LLM communication, making it easy to switch providers: 
// Switch to Anthropic's Claude
const agent = new Agent({
  name: 'math-assistant',
  provider: new Anthropic({ 
    apiKey: 'your-api-key', 
    model: 'claude-3-sonnet-20240229' 
  }),
  tools: [new Calculator()],
  instructions: 'You are a helpful math assistant.',
});

// Or use a locally-hosted model with Ollama
const agent = new Agent({
  name: 'math-assistant',
  provider: new Ollama({ 
    model: 'llama3.2' 
  }),
  tools: [new Calculator()],
  instructions: 'You are a helpful math assistant.',
});

​
Creating custom providers

You can create custom providers by implementing the LLMProvider interface. This allows you to:
  1. Support proprietary or internal LLM services
  2. Add support for new public LLM services
  3. Create advanced wrappers around existing providers
Here’s a simplified example of implementing a custom provider: 
import { BaseLLMProvider, GenerateOptions, GenerateResponse, PartialReturn } from '@agenite/llm';

class CustomProvider extends BaseLLMProvider {
  name = 'custom-provider';
  version = '1.0.0';
  
  constructor(private config: CustomConfig) {
    super();
  }

  async generate(
    input: string | BaseMessage[],
    options?: Partial<GenerateOptions>
  ): Promise<GenerateResponse> {
    // Convert Agenite input format to your provider's format
    // Make API call to your LLM service
    // Convert provider response to Agenite GenerateResponse format
    // Handle errors appropriately
  }

  async *stream(
    input: string | BaseMessage[],
    options?: Partial<GenerateOptions>
  ): AsyncGenerator<PartialReturn, GenerateResponse, unknown> {
    // Similar to generate, but with streaming implementation
  }
}
By extending BaseLLMProvider, you only need to implement the generate and stream methods - the iterate method will be handled for you.

​
Provider best practices

When working with providers, follow these best practices:
  1. Store API keys securely: Never hardcode API keys or credentials. Use environment variables or secure secret management.
  2. Handle rate limits: Implement appropriate retry logic and respects rate limits of the LLM service.
  3. Implement timeouts: Set reasonable timeouts to handle slow or non-responsive API calls.
  4. Configure for your use case: Adjust provider configuration options to match your specific use case:
    • Lower temperatures (0.0-0.3) for more deterministic responses
    • Higher temperatures (0.7-1.0) for more creative responses
    • Appropriate max tokens based on expected response length
  5. Implement error handling: Always handle errors from the LLM service gracefully. 
try {
  const result = await agent.execute({
    messages: [{ role: 'user', content: userQuery }]
  });
  // Process result
} catch (error) {
  if (error.code === 'RATE_LIMIT_EXCEEDED') {
    // Implement backoff strategy
  } else if (error.code === 'CONTEXT_LENGTH_EXCEEDED') {
    // Handle too much input
  } else {
    // Handle other errors
  }
}

​
Conclusion

Providers are a key component of Agenite’s flexibility. By implementing a common interface for different LLM services, providers enable your agents to use the best model for the job without being locked into a single vendor or technology. In the next section, we’ll explore the concept of tools in Agenite and how they extend the capabilities of your agents.