Gentopia-Mason

IMPORTANT NOTICE: This code repository was adapted from Gentopia.AI to support Mason Activities.

Authors: Ziyu Yao ([email protected]), Saurabh Srivastava ([email protected]), and Murong Yue ([email protected])

Copyright and license should go to Gentopia.AI.

Windows Users: We currently only support Powershell.

Installation 💻

First, clone this repository:

git clone [email protected]:LittleYUYU/Gentopia-Mason.git
cd Gentopia-Mason

If you have not set up your ssh keys yet, you may receive an error. Instead, you can clone the repository using the following command:

git clone https://github.com/LittleYUYU/Gentopia-Mason.git

Next, we will create a virtual environment and install the library:

conda create --name gentenv python=3.10
conda activate gentenv
pip install -r requirements.txt

Most of the agent construction and execution activities will happen within GentPool. For the gentopia library to be used within GentPool, set up the global environment:

For Linux/Mac

export PYTHONPATH="$PWD/Gentopia:$PYTHONPATH"

For Windows (Windows Powershell):

$env:PYTHONPATH = "$PWD/Gentopia;$env:PYTHONPATH"

In addition, since we will be using OpenAI's API, we also need to create a .env file under GentPool and put the API Key inside. The key will be registered as environmental variables at run time. To do this, first, let's change the directory:

cd GentPool

For Linux/Mac

touch .env
echo "OPENAI_API_KEY=<your_openai_api_key>" >> .env

For Windows (Windows Powershell)

$env:OPENAI_API_KEY="<YOUR KEY>"

Now you are all set! Let's create your first Gentopia Agent.

Quick Start: Clone a Vanilla LLM Agent

GentPool has provided multiple template LLM agents. To get started, we will clone the "vanilla agent" from GentPool/gentpool/pool/vanilla_template with the following command:

For Linux/Mac

./clone_agent vanilla_template <your_agent_name> 

For Windows (Windows Powershell)

.\clone_agent.bat vanilla_template <your_agent_name> #note the .bat file 

This command will initiate an agent template under ./GentPool/gentpool/pool/<your_agent_name>. The agent configuration can be found in ./GentPool/gentpool/pool/<your_agent_name>/agent.yaml (note the agent type vanilla). The vanilla prompt it uses can be found in the source code of Gentopia; see ./Gentopia/gentopia/prompt/vanilla.py.

You can now run your agent via:

python assemble.py <your_agent_name>

This vanilla agent simply sends the received user query to the backend LLM and returns its output. Therefore, for many complicated tasks, such as those requiring accessing the latest materials, it will fail.

Implement a Scholar LLM Agent with Tool Augmentation

In the second trial, we will create a Scholar agent which is augmented with multiple functions to access Google Scholar in real time.

This is based on the scholar agent we have created in the pool. As before, in this demo we simply clone it:

For Linux/Mac

./clone_agent scholar <your_agent_name> 

For Windows (Windows Powershell)

.\clone_agent.bat scholar <your_agent_name> #note the .bat file 

Like before, this command created an agent under ./GentPool/gentpool/pool/<your_agent_name>. Note from its configuration that scholar is an openai-type agent. As stated in its Gentopia's implementation, this type of agent allows for function calling:

OpenAIFunctionChatAgent class inherited from BaseAgent. Implementing OpenAI function call api as agent.

The available functions to the scholar agent have been listed in its configuration file ./GentPool/gentpool/pool/<your_agent_name>/agent.yaml, and the implementation of these tools can be found in Gentopia's source code (mostly coming from the google_scholar.py file, in this example).

Now, you are all set to query this scholar agent for the latest papers by certain authors, the summary of a certain paper, paper citations, etc.