AgentX

AgentX is an experiment to develop an autonomous agent that delegates well to Auto-GPT, babyagi, and other agents using LangChain. It is a LangChain-powered agent that delegates tasks to other AI agents until user's objective is complete. It aims to assist users in completing objectives using agent retrieval, whereby objectives are outsources to other AI agents (Auto-GPT, babyagi, HuggingFace) in parallel.

Requirements

AgentX requires Python 3.8 and Poetry for package management.

Set Up

Here's how to install AgentX:

1. Clone the repository:

git clone [email protected]:slavakurilyak/agentx.git
cd agentx

2. Set up the project and activate the virtual environment via Poetry:

poetry install
source $(poetry env info --path)/bin/activate

3. Copy .env.example to .env:

cp .env.example .env

4. Update your OPENAI_API_KEY and SERPAPI_API_KEY values inside newly created .env

Usage

To run the 'experimental version' of AgentX using Python, type:

python agentx/agents/autogpt/main.py
python agentx/agents/babyagi/main.py

-or-

poetry run python agentx/agents/autogpt/main.py
poetry run python agentx/agents/babyagi/main.py

To run the 'stable version' of AgentX using Python, type:

python agentx/autogpt.py
python agentx/baby_agi_with_tools.py

-or-

poetry run python agentx/autogpt.py
poetry run python agentx/baby_agi_with_tools.py

Contibuting

We use black, isort and flake8 to ensure good code quality.

Always run the linters to check your code using:

isort --check .
black --check .
flake8

-or-

poetry run isort --check .
poetry run black --check .
poetry run flake8

To use isort and black for autoformatting, run the commands without the --check option using:

isort .
black .
flake8

-or-

poetry run isort .
poetry run black .
poetry run flake8

Run tests in the root directory either using:

pytest

-or-

poetry run pytest

How does AgentX work?

1. User requests assistance from AgentX.
2. AgentX retrieves necessary tools.
3. AgentX selects relevant tools.
4. Selected tools complete the task and their results are combined.
5. AgentX processes the combined results and presents the response to the user.

This user-AgentX interaction can be visualized as:

graph TB
A[User] --> B[AgentX]
B --> C[Find Relevant Agent]
C --> D{Delegate Task to Agent}
D --> E1[Auto-GPT]
D --> E2[BabyAGI]
D --> E3[HuggingFace]
D --> E4[Other AI Agent]
E1 --> F1[Complete Task with Auto-GPT]
E2 --> F2[Complete Task with BabyAGI]
E3 --> F3[Complete Task with HuggingFace]
E4 --> F4[Complete Task with Other AI Agent]
F1 --> G[Evaluate Task Completion]
F2 --> G
F3 --> G
F4 --> G
G --> H[AgentX]
H --> I[User]

Loading

Benefits of using AgentX

• Delegates tasks to multiple AI agents for efficient handling of complex tasks.
• Can handle a wide range of tasks, from answering questions to creating to-do lists.
• Can be customized to suit your specific needs.

Limitations of AgentX

At present, AgentX is limited in its ability to interact with Auto-GPT and babyagi directly due to the following reasons:

• These models do not currently have APIs available for interaction.
• They also lack JSON manifest files such as .well-known/ai-plugin.json and .well-known/openai.yaml, which are required by OpenAI.

As a result of these limitations, AgentX can only interact with Auto-GPT and babyagi by directly cloning their source code using $ git clone. We will provide further information on these limitations shortly.

Progress

• BabyAGI using LangChain
• Auto-GPT using LangChain
• Auto-GPT from source (see issue)
• babyagi from source (see issue)
• Teenage-AGI from source (see issue)

Sure, here's the section about contributions are welcome:

Get involved

We appreciate your interest in AgentX! Here are some ways you can get involved:

• If you like our project, please consider starring it on GitHub.
• If you have an idea for a new feature or improvement, please open a new issue on GitHub.
• If you see a bug or want to implement a feature, please open a new pull request on GitHub.
• If you want to follow the founder on Twitter, visit twitter.com/slavakurilyak.

Thank you for your support!

Star History