LLM Adapters Core

Unified interface for multiple AI model providers.

Architecture

AbstractClient

Base class in src/core/adapters/AbstractClient.js:

class AbstractClient {
  // Send message to LLM
  async sendMessage(messages, options) { }
  
  // Parse tool calls from response
  parseToolCalls(response) { }
  
  // Convert message format
  convertMessages(messages) { }
}

Supported Providers

Provider	Client	Features
OpenAI	OpenAIClient	Tool calling, streaming, vision
Claude	ClaudeClient	Tool use, streaming
Gemini	GeminiClient	Function calling
DeepSeek	Uses OpenAI client	OpenAI-compatible

Message Flow

Creating Custom Adapter

1. Extend AbstractClient
2. Implement required methods
3. Register in adapter factory

import { AbstractClient } from './AbstractClient.js'

export class MyClient extends AbstractClient {
  async sendMessage(messages, options) {
    // Implementation
  }
}

Next Steps

• Chat Service - Message handling
• MCP System - Tool integration