LLM Adapters

This directory contains a collection of adapter classes that provide a standardized interface for interacting with various Large Language Model (LLM) providers. Each adapter implements the BaseVendorAdapter abstract base class to ensure consistent handling of model interactions across different vendors.

Overview

These adapters enable our system to work seamlessly with multiple LLM providers by converting vendor-specific APIs into a unified interface. All adapters produce standardized Server-Sent Events (SSE) chunks when streaming text responses.

Available Adapters

Adapter	Description
BaseVendorAdapter	Interface all adapters must implement
AnthropicAdapter	Adapter for Anthropic's Claude models
OpenAIAdapter	Adapter for OpenAI models
OpenAICompatAdapter	Adapter for APIs compatible with OpenAI's interface
MistralAIAdapter	Adapter for Mistral AI models
XAIAdapter	Adapter for xAI models
WatsonxAdapter	Adapter for IBM's watsonx.ai platform

WatsonX Submodule

The watsonx directory contains specialized implementations for IBM's watsonx.ai platform:

• WatsonxAdapter - Main adapter for watsonx.ai models
• IBMTokenManager - Handles authentication with IBM Cloud
• WatsonxConfig - Configuration for watsonx.ai connections

Core Features

All adapters implement these key methods:

• gen_sse_stream(prompt: str) - Generate streaming responses from a text prompt
• gen_chat_sse_stream(messages: List[TextChatMessage], tools: Optional[List[Tool]]) - Generate streaming responses from a chat context

Implementation Requirements

When implementing a new adapter:

1. Inherit from BaseVendorAdapter
2. Implement all abstract methods
3. Convert vendor-specific responses to our standardized SSEChunk format
4. Handle streaming appropriately for the specific vendor API

Usage Example

# Example of using an adapter
from src.llm.adapters import OpenAIAdapter

# Initialize with model name and default parameters
adapter = OpenAIAdapter(
    model_name="gpt-4o-mini",
    temperature=0.7,
    max_tokens=1000
)

# Generate streaming response
async for chunk in adapter.gen_sse_stream("Tell me about machine learning"):
    # Process each SSEChunk
    if chunk.choices and chunk.choices[0].delta.content:
        content = chunk.choices[0].delta.content
        # Process the content fragment
        print(content, end="")

Adding New Adapters

To add support for a new LLM provider:

1. Create a new file named your_provider_adapter.py
2. Implement the BaseVendorAdapter interface
3. Add docstrings and logging
4. Update this index with your new adapter

---