Custom Conversable Agent

This repository hosts two versions of a custom conversable agent named "Lou the Corgi". These agents are built using the AutoGen framework, enriched with unique identity prompts, and integrate a teachability module to adapt and learn from interactions. The advanced version extends functionality with self-reflection capabilities and direct API integration.

Features:

Custom Identity Prompt: Both versions of the agent enhance user experience by responding with a unique identity.
Teachability: Integrates a learning component allowing the agent to improve based on past interactions, present in both versions.
User Proxy Agent: Simulates user interactions, useful for testing and development, used in both versions.
Self-Reflection and API Interaction: Exclusive to the advanced version, enabling the agent to reflect on its thoughts and interact directly with APIs.

Requirements

Ensure Python 3.8+ is installed, then install dependencies from requirements.txt:

autogen==1.0.16
virtualenv
requests  # Required for the advanced version

Installation

Clone this repository:
```
git clone <repository-url>
```
Navigate to the cloned directory:
```
cd <repository-directory>
```
Install the required Python packages:
```
pip install -r requirements.txt
```

Usage

Run the agent by executing the Python script:

python custom_conversable_agent.py

Code Explanation

Environment Configuration

Disabling tokenizer parallelism to avoid multi-threading issues:

import os
os.environ["TOKENIZERS_PARALLELISM"] = "false"

Advanced Custom Conversable Agent Class

Enhancements include API interactions and added methods for self-reflection:

from autogen import ConversableAgent
import requests

class CustomConversableAgent(ConversableAgent):
    # Initialization includes storing API keys and base URL
    def __init__(self, name, llm_config, identity_prompts, *args, **kwargs):
        ...
    
    # Enhanced request handling to include self-reflection and API requests
    def handle_request(self, message):
        ...
    
    # Self-reflection method
    def reflect(self, message):
        ...
    
    # API interaction method
    def llm_predict(self, prompt):
        ...

Teachability

Enabling the agent to learn and adapt from user interactions is consistent across both versions:

from autogen.agentchat.contrib.capabilities.teachability import Teachability

teachability = Teachability(
    reset_db=False,
    path_to_db_dir="./tmp/interactive/teachability_db"
)

Initiation Function

Starts the chat with a welcoming message:

def initiate_chat():
    ...

2acrestudios/autogen-teachable-agent