This module provides a unified and extensible interface for interacting with multiple large language model (LLM) providers, enabling consistent and modular event handling for AI-driven applications. It follows an object-oriented design to support OpenAI, Gemini, Anthropic, and Ollama models, all under a shared abstraction layer: AIAgentEventHandler.
At the core of the system is the AIAgentEventHandler base class, which defines shared logic and interface methods for invoking model functions, streaming outputs, and handling responses.
Four concrete handler classes inherit from this base:
OpenAIEventHandler: Interfaces with OpenAI's models.GeminiEventHandler: Connects to Google's Gemini API through thegenai.Client.AnthropicEventHandler: Manages interactions with Anthropic's Claude models.OllamaEventHandler: Handles local LLM inference with Ollama and supports tool usage.
Each subclass is responsible for:
- Managing model-specific clients and configuration.
- Implementing
invoke_model()and optionallystream_response()methods. - Managing session memory, threading, and streaming tokens.
- ✅ Unified API for LLM invocation
- 🔁 Support for both standard and streaming responses
- 🔌 Tool usage and system prompts for advanced interactions (Ollama)
- 🧠 Assistant memory support for context-aware interactions
- 📊 Compatible with modular function routing and observability layers
This module is designed for applications that orchestrate LLMs across different providers—such as chatbots, agent frameworks, RAG pipelines, or AI-driven workflows—while maintaining a consistent interface and centralized control.