With the explosion of ChatGPT over the internet, many people were impressed by the power and utility of OpenAI's GPT technology. This sparked the idea of making an automatic ChatGPT that answers and generates its own prompts to achieve a specific goal. This idea would later be carried out as Auto-GPT.

Since Auto-GPT is still under development, you'll only be able to access Auto-GPT just how a developer would—which may require a bit of technical know-how.

To make things easier for you, here is a step-by-step guide on how to download and install Auto-GPT.

Step 1: Download Python and AutoGPT

Despite what you may have read elsewhere, installing Auto-GPT is pretty straightforward.

Let's begin by manually downloading the latest version of Python 3 and the Auto-GPT executable from GitHub. You'll first want to download and install Python 3 since your PC will need it to read and execute files within Auto-GPT.

Download: Python 3 (Free)

Once downloaded, double-click on the file to install Python. Make sure to tick the box for Add python.exe to PATH. This will enable your PC to use Python anywhere in your PC. After that, go ahead and click Install Now.

A tab showing a tab to install Python

After Installing Python, you can download Auto-GPT from GitHub.

Download: Auto-GPT (Free)

Source code.zip is for Windows, while Source code.tar.gz is for Linux and MacOS. First, download the file for your operating system, then copy the folder and paste it into your desired location.

Copy AutoGPT to destination folder

Step 2: Configure Auto-GPT

Since AutoGPT uses OpenAI's GPT technology, you must generate an API key from OpenAI to act as your credential to use their product.

Keep in mind that your account on ChatGPT is different from an OpenAI account. You will need to register for an OpenAI account to access an OpenAI API. Now:

  1. After registration and login, click on Personal in the top right corner of the website and select View API keys. This will send you to the OpenAI API keys management, where you can manage your API keys.
  2. To create a key, click Create new secret key, input a name, then click Create secret key. You can then copy the API key by using CTRL + C or clicking the copy icon on the right.
    Create API key
  3. Now that you have your API key, go to your Auto-GPT folder and open the .env file using Notepad.
    Open env with Notepad
  4. Once opened, scroll down to the LLM PROVIDER section. There you will see OPENAI_API_KEY. Replace the placeholder with the API key you've just copied, then save the file.
    Replace API key

This file is where all your service credentials are placed, so if you want to use a backend vector database like Docker or Pinecone, you can set your product API keys here. But if you only want to use AutoGPT, the OpenAI API key should be enough.

Step 3: Install Auto-GPT Dependencies

Now that you have configured Auto-GPT, it's time to install its dependencies through a terminal.

  1. To open a terminal in the Auto-GPT environment, right-click on the Auto-GPT folder, then select Open in Terminal.
    Open AutoGPT environment
  2. To install all the requirements needed for Auto-GPT to work, use the command:
    pip install -r requirements.txt
  3. Once you press enter, your terminal will download and install all the required dependencies.
    Pip install requirements
  4. After installation, try opening Auto-GPT by using:
    python -m autogpt
    AutoGPT installation success

Congratulations! You have successfully Installed Auto-GPT.

How to Use Auto-GPT

Now that you have successfully installed Auto-GPT on your computer, let's discuss how to use Auto-GPT.

When you first open Auto-GPT, you'll immediately be asked to give a name for your AI assistant, its role, and five goals the AI should follow.

You can input any name you want. It likely wouldn't affect the performance of your AI assistant. After giving a name, try providing a clear and concise role, as this will set the role the AI will play.

Although you don't need to fill all five goals, it is still recommended that you do, as this will likely affect the efficiency of your AI.

Creating Recipe-Generator

In this example, we have named our AI assistant "Recipe-Generator." Its role is to make a recipe based on the top five ingredients readily available in the US. We've set the first three goals as parameters on what we expect the recipe will be and set the last two to tell Auto-GPT to save the file as TXT, then shutdown.

Running Recipe-Generator

Once you give your last goal, you can hit enter for Auto-GPT to run.

While running, you can see the AI's thoughts, reasoning, plan, and criticism. For every action of the AI assistant, you will be asked to authorize its plan to execute. You can do so by typing "y" as yes.

If you want the AI to continue a number of times without asking you for authorization, you can type "y -(number actions authorized)." For example, if you want your AI assistant to continue executing the following five steps, you can type "y -5" and hit enter.

One advantage of Auto-GPT over ChatGPT is that it is free to probe through the internet. As you can see here, our Recipe-Generator assistant downloads a file.

AutoGPT downloading file

This makes this AI potentially dangerous; that's why Auto-GPT always asks you for authorization before executing plans. Always read and understand your AI assistant's thoughts, reasoning, and plan before authorizing its actions.

After every action of the AI, you can also provide your feedback to help the AI with its task.

Providing human input

In this screenshot, our AI assistant has looped through the same step three times. So, we tell the AI to skip browsing for recipes and start creating the output.

Recipe-Generator success

After making the recipe, our AI has now completed its task.

To view the output, go to your Auto-GPT folder and open auto-gpt-workspace.

Viewing AutoGPT output

Success! Our AI assistant has given us a recipe for a chicken pot pie casserole.

Auto-GPT Limitations

Although it was nice that Auto-GPT is working, it's still quite limited. Through a series of testing, we discovered that Auto-GPT couldn't handle anything complex. Most of the time, it continued looping around the same thought and reasoning. Although you can definitely keep providing helpful prompts, it would feel more like ChatGPT instead of the autonomous assistant it was meant to be.

Many of these endless loopings were caused by the AI not knowing how to handle or navigate through website ads, log-in pages, cookies, and all kinds of pop-ups. This difficulty was likely caused by the fact that we were only running Auto-GPT on GPT-3.5 instead of GPT-4.

Although the two versions may be close in terms of version, GPT3.5 and GPT-4 are very different from one another.

The Future of Auto-GPT

There are two reasons why Auto-GPT isn't as powerful as it should be.

The first is that it's still in development—more features and tweaking must be added to make Auto-GPT function as intended. The second is that GPT-3.5 wasn't meant to power Auto-GPT. If you look at the official GitHub, you can see that AutoGPT is being developed to use GPT-4.

It will only be a matter of time before AutoGPT gets to a mature stage of development and GPT-4 is released. When that happens, AutoGPT will likely be as impressive and powerful as it was first thought to be. So, before that happens, you may be interested in playing with it while we wait for its upcoming updates and GPT-4.