Autogen with Chroma DB

AutoGen is an open-source framework that enables the development of conversational AI applications using multiple agents.

Chroma DB is an open-source vector database for storing and retrieving vector embeddings.

Create virtual python environment

• virtualenv -p python3.11 env_name
• python -m venv env_name

Activate the virtual env

• env_name/scripts/activate

Installs AutoGen & Chroms DB

pip install -U "pyautogen[retrievechat]" chromadb

• -U tells pip to upgrade any already installed packages to their latest versions before installing.
• "pyautogen[retrievechat]" installs the pyautogen package and also installs the optional "retrievechat" extra feature of that package

Set environment variable AUTOGEN_USE_DOCKER to False

Bash Command:

export AUTOGEN_USE_DOCKER=False

PowerShell Command:

$Env:AUTOGEN_USE_DOCKER="False"

Exporting AUTOGEN_USE_DOCKER=False tells pyautogen to run its tasks directly on the host rather than using Docker containers. It bypasses the Docker dependency but also loses some of the isolation benefits Docker provides.

Set environment variable OPENAI_API_KEY=???

Bash Command:

export OPENAI_API_KEY=Fxxxxxxxxxxxxxxxxxxxxxxxxx

PowerShell Command:

$Env:OPENAI_API_KEY="xxxxxxxxxxxxxxxxxxxxxxxxx"

Run `app.py`

python app.py

Explanation of the code file

This code file defines a chatbot system using the autogen and chromadb libraries. Here's a step-by-step breakdown of the code:

Importing Libraries

The first step is to import the necessary libraries. In this case, we're using autogen and chromadb to create a chatbot that can retrieve information from a database and generate responses based on a language model.

import autogen
import chromadb

Defining the Chatbot Assistant

Next, we define the chatbot assistant using the AssistantAgent class from the autogen library. This class takes a name, language model configuration, and system message as input.

assistant = AssistantAgent(
    name="my_assistant",
    llm_config=llm_config_proxy,
    system_message="You are a helpful assistant. Provide accurate answers based on the context. Respond 'Unsure about answer' if uncertain."
)

Defining the User

We also define the user using the RetrieveUserProxyAgent class from the autogen.agentchat.contrib module. This class takes a name, human input mode, system message, maximum number of consecutive auto-replies, and configuration for retrieving information from a database as input.

user = RetrieveUserProxyAgent(
    name="me_user",
    human_input_mode="NEVER",
    system_message="Assistant who has extra content retrieval power for solving difficult problems.",
    max_consecutive_auto_reply=10,
    retrieve_config={
        "task": "code",
        "docs_path": ['./docs/autogen.pdf'],
        "chunk_token_size": 1000,
        "model": config_list[0]["model"],
        "client": chromadb.PersistentClient(path='/tmp/chromadb'),
        "collection_name": "pdfreader",
        "get_or_create": True,
    },
    code_execution_config={"work_dir": "coding"},
)

Defining the User Question

We define the user's question or prompt as a string variable.

user_question = """
Compose a short blog post showcasing how AutoGen is revolutionizing the future of Generative AI 
through the collaboration of various agents. Craft an introduction, main body, and a compelling 
conclusion. Encourage readers to share the post. Keep the post under 500 words.
"""

Initiating the Chat

Finally, we initiate the chat session between the user and the chatbot using the initiate_chat method of the RetrieveUserProxyAgent class.

user.initiate_chat(assistant, problem=user_question)

Summary

Overall, this code file defines a chatbot system that can respond to user questions or prompts by retrieving information from a database and generating responses based on a language model. The chatbot can also execute code and provide answers based on the context of the user's question.

Links

• Getting started with Chroma DB
• AutoGen: Enable Next-Gen Large Language Model Applications

🚀 Quick Start

pip install -U "pyautogen[retrievechat]" chromadb