An Overview Of Autonomous Agents: BabyAGI, Auto-GPT, CAMEL, and Beyond
A Guide to the Latest Open-Source Innovations in AI-Powered Autonomous Agent Frameworks
The interest in tools like BabyAGI and Auto-GPT is tremendous. Has there ever been an open-source project that has been catapulted into the focus of both developers and creatives like this?
The speed at which new versions and variants of these frameworks are coming out is breathtaking. So I guess, it’s time for us to catch up with the latest releases and look at how these agents work, what types we have so far, and how you can try them yourself.
In this post, we’ll have a look at
- BabyAGI
- Auto-GPT
- AgentGPT
- Godmode
- Do Anything Machine
- Microsoft’s JARVIS or HuggingGPT
- AI Legion
- CAMEL
- GPTRPG
(I know this is not a complete list. Feel free to tell me in the comments what’s missing, and I’ll add it.)
BabyAGI
On April 3rd, Yohei Nakajima open-sourced a project he had been working on to automate his personal task management procedure.
The ironically named BabyAGI was set free!
The script behind BabyAGI is surprisingly simple (don’t let the seemingly complicated diagram fool you). It’s essentially a language model interacting with a task list, with the objective of automatically generating, prioritizing and executing tasks based on a predefined goal.
The completed tasks are stored in a database and act as the language model’s “memory,” from which it can derive context for the creation and execution of new tasks. The basic BabyAGI script has no modules for internet research or running code autonomously. However, those updates have been implemented in spin-off versions. Check out Yohei’s list of projects that build upon BabyAGI: https://github.com/yoheinakajima/babyagi/blob/main/inspired-projects.md
Website: http://babyagi.org/
Twitter: https://twitter.com/babyAGI_
Github: https://github.com/yoheinakajima/babyagi
Auto-GPT
Take a look at the history of people showing their enthusiasm for this open-source project on GitHub:
Auto-GPT is an experimental open-source project that uses GPT-4 to concatenate AI “thoughts” (i.e., the model thinks about, criticizes, and re-evaluates tasks) and autonomously tries to achieve the goal you specify. Auto-GPT pushes the boundaries of what is possible with AI by giving the model the ability to “execute commands,” meaning it can decide which tools from a given arsenal to use and how to use them. (here’s a quick tutorial on how to setup & run AutoGPT locally)
AutoGPT can scrape websites, search for information, generate images, create and run code, etc. Here’s a list of current commands it’s able to run:
GitHub: https://github.com/Significant-Gravitas/Auto-GPT
Run it on Replit: https://twitter.com/nathanwchan/status/1646194631577833472?s=20
Setup Tutorial: https://readmedium.com/complete-guide-to-setup-autogpt-revolutionize-your-task-automation-with-gpt-4-39eda5a85821
AgentGPT
AgentGPT brings the idea of Auto-GPT/BabyAGI to the web. You can deploy your own Autonomous Agent via your browser here: https://agentgpt.reworkd.ai/
The following capabilities have been added to AgentGPT at the time of writing:
- Long-term memory via a database
- Web browsing Interaction with websites and people
- Saving agent runs
Website: https://agentgpt.reworkd.ai/
Twitter: https://twitter.com/asimdotshrestha/status/1644883727707959296
Github: https://github.com/reworkd/AgentGPT
Godmode
Speaking of web apps inspired by Auto-GPT, there is another one you might want to try: Godmode. This app also provides users with an interactive and user-friendly interface, enabling them to utilize Autonomous Agents to solve tasks.
Do Anything Machine
Another web-based AutoGPT-like project is the “Do Anything Machine.” There is currently a waitlist to apply for early access to this highly sophisticated project that aims to have multiple Autonomous Agents working on tasks in parallel, all while having access to your apps and background information.
Website: https://www.doanythingmachine.com/
Microsoft’s JARVIS
Microsoft JARVIS, also known as HuggingGPT, is a collaborative system that uses multiple AI models to accomplish a given task, with OpenAI’s GPT models acting as the controller.
JARVIS is integrating various open-source models for images, videos, audio, etc., and can also connect to the internet and access files. In a similar manner to BabyAGI & AutoGPT, JARVIS will analyze the tasks, as well as select the appropriate model to complete them.
App: https://huggingface.co/spaces/microsoft/HuggingGPT
Github: https://github.com/microsoft/JARVIS
AI Legion
“A framework for autonomous agents who can work together to accomplish tasks.”
This is another interesting approach where you interact with several agents at once through the console where multiple AI agents collaborate and communicate with you to complete tasks more effectively.
Github: https://github.com/eumemic/ai-legion
CAMEL
CAMEL, short for “Communicative Agents for ‘Mind’ Exploration of Large Scale Language Models,” is essentially a role-playing approach to the loop architecture of BabyAGI & AutoGPT.
With CAMEL, you have two agents to which you assign specific roles. Then, you just watch them as they collaborate and work together to solve your task.
Web Demo: http://agents.camel-ai.org/
Project Page: https://www.camel-ai.org/
Github: https://github.com/lightaime/camel
GPTRPG
Finally, a gamified approach to the autonomous agents paradigm can be found in this experimental repo on GitHub: https://github.com/dzoba/gptrpg
The repo consists of two main components:
- A simple RPG-like environment for an LLM-enabled AI Agent to exist in
- A simple AI Agent connected to the OpenAI API to exist in that environment
This is building on a recently released paper in which multiple agents have been deployed to autonomously participate in an online game. It’s a fascinating concept! You can check out the paper and learn more about the research here: https://arxiv.org/abs/2304.03442
➡️ For more information about AI & Creativity, follow me on Twitter or Medium (use my referral link to get full access to all my articles and those of thousands of other writers).
➡️ If you like my content, why not leave a “clap” at the end of this article, so more people can see it?
Stay updated with the latest news and updates in the creative AI space — follow the Generative AI publication.
