Allowing Local Execution Without Docker

[teamcoltra]

I recognize that this largely just stems from my ignorance of docker but it does feel like every time I try to do things with Docker it's more hassle than it's worth. I am already running AutoGPT in a virtual machine so from my end it's safe and can't really do a whole lot of damage but I would really love it to be able to live outside of docker and freely interact with the files that are on the virtual machine while also doing code execution.

Every time anything wants to execute it says that it can only execute in docker mode. Is there a setting to disable that?

[sohrabsaran]

Docker usage is a great idea to prevent unexpected damage by the Autonomous AI's actions, to the local filesystem.
I'm also trying to get up to speed with Docker and had contributed a fix to Auto-GPT in this area recently.
One advantage that Docker has, is that it is significantly more lightweight and performant than a VM as per my understanding.
So it is worth learning and seems here to stay.
If you can please share the exact issue you face, then perhaps me or someone else can push this forward.

[LanceLake]

Docker usage is a great idea to prevent unexpected damage by the Autonomous AI's actions, to the local filesystem.

He is aware of this as he uses a VM.

If you can please share the exact issue you face, then perhaps me or someone else can push this forward.

He wants to turn off the Docker setting. As do I.

So let me give you an exact issue.. I don't know how to use docker at all and it's not easy to set up. I have it and try to run it, but I need to edit the .env file, but I can't find it in the files.

Perhaps in your fixes that you are doing, you can make it so that it can be turned off for those who don't know how to use docker?

[sohrabsaran]

I am talking about the 'without docker as far as possible' instructions at the page at https://docs.agpt.co/setup/
That is more in line with the subject of this discussion thread.
I am not sure what you mean by 'within the docker'.
You can search for the following on this page:

Set up without Git/Docker

This says:

1. Download Source code (zip) from the [latest stable release](https://github.com/Significant-Gravitas/Auto-GPT/releases/latest)
2. Extract the zip-file into a folder

Here is another way to set it up, that I used. Rather than download the zip file, you can do the following:
Run these commands in a new folder, perhaps your old folder is messed up somehow:

git clone https://github.com/Significant-Gravitas/Auto-GPT.git
cd Auto-GPT
git checkout stable
cat .env.template

This clones the repository into the default folder Auto-GPT, changes directory to the folder, checks out the 'stable' branch and displays the content of the .env.template file.
Note: this should be common knowledge but just in case: depending on the preferences set on your machine, windows explorer (or equivalent file managers on other operating systems) may not display files whose name start with '.' by default.

Note2: Though the above method of installation is called 'installation without docker', the execute_python_file command (that AutoGPT may run depending on the goals set in AutoGPT) downloads and runs a small docker image that has python installed in it. To be precise, it transfers into this small docker image, a python file that has been generated by AutoGPT based on the goals set in AutoGPT, and then executes that python file in that docker image. This small docker image acts as a security sandbox to prevent reads/writes to the host file system, that could have unforeseen disastrous effects such as data loss, data corruption, exposure of private data etc. as many users may not think of taking the precaution to run AutoGPT in a VM that can afford to be corrupted or have its data exposed. For this reason, the machine on which AutoGPT is cloned to, must also have docker installed on it, so that the execute_python_file command can work.

[LanceLake]

It was discussing not requiring docker. You came in and said it was unsafe and it should run within docker. I pointed out that docker is difficult to set up and run the AI within it. That's how the conversation went. I'm sorry if you got confused as to what the issue was that you were arguing against. All we would like is to not have to require docker to run python scripts.

To be precise, it transfers into this small docker image, a python file that has been generated by AutoGPT based on the goals set in AutoGPT, and then executes that python file in that docker image. This small docker image acts as a security sandbox to prevent reads/writes to the host file system, that could have unforeseen disastrous effects such as data loss, data corruption, exposure of private data etc. as many users may not think of taking the precaution to run AutoGPT in a VM that can afford to be corrupted or have its data exposed. For this reason, the machine on which AutoGPT is cloned to, must also have docker installed on it, so that the execute_python_file command can work.

Ah.. Then there is an issue there. It doesn't work.

I have docker installed. I get this error when it tries to execute any python file and docker is running.

SYSTEM: Human feedback: Write a python script to do something
JESSICA RABBIT THOUGHTS: The user requested for a python script to be written. Since the task specifics were not provided, I will create a simple script that prints 'Hello, World!'. This will be done using the 'execute_python_code' command.
REASONING: A Python script needs to be written as per the user's request. The task specifics were not provided, so a simple script will be created.
PLAN:

• Create a Python script
• Use the 'execute_python_code' command to create and execute the script
  CRITICISM: There was not enough information provided to create a more complex or tailored script.
  SPEAK: Creating a Python script that prints 'Hello, World!'.

NEXT ACTION: COMMAND = execute_python_code ARGUMENTS = {'code': "print('Hello, World!')", 'name': 'hello_world.py'}
Enter 'y' to authorise command, 'y -N' to run N continuous commands, 'n' to exit program, or enter feedback for Jessica Rabbit...
Asking user via keyboard...
Waiting for your response...y
-=-=-=-=-=-=-= COMMAND AUTHORISED BY USER -=-=-=-=-=-=-=
Executing python file 'C:\Coderoot\AI\AutoGPT\Pure\autogpt\workspace\auto_gpt_workspace\Jessica Rabbit\executed_code\hello_world.py' in working directory 'C:\Coderoot\AI\AutoGPT\Pure\autogpt\workspace\auto_gpt_workspace'
Could not run the script in a container. If you haven't already, please install Docker https://docs.docker.com/get-docker/
SYSTEM: Command execute_python_code returned: Error: Error while fetching server API version: (2, 'CreateFile', 'The system cannot find the file specified.')

[sohrabsaran]

Looks like the .env.template issues is resolved. 'Jessica Rabbit' has a space in it. Can you please try a path without spaces by choosing a one word name for your AI agent? Sorry I'm unable to help more right now as am caught up in other things. execute_python_file works for me, so we'll get it working for you no worries.

[LanceLake]

This also happens with the name of "Jessica". It's not the username.

[sohrabsaran]

Do you have Docker installed locally?
If yes, we'll need to figure out where this exception is coming from. Looks like once we do that, we can improve the error handling of the codebase. Search for this in the code:

    except DockerException as e:
        logger.warn(
            "Could not run the script in a container. If you haven't already, please install Docker https://docs.docker.com/get-docker/"
        )
        return f"Error: {str(e)}"

...and before the return statement, log the exception 'e', refer https://www.geeksforgeeks.org/how-to-print-exception-stack-trace-in-python/
...and share the stack trace.