Isolate VMs and Networks

[significant-gravitas]

Isolate VMs and Networks

[slavakurilyak]

Similar to how the OpenAI platform shows usage, Github Codespaces shows similar analytics.

Here's an example of GitHub Codespaces usage:

Github provides a way to manage Codespaces user secrets via REST API without revealing their encrypted values.

GitHub's secrets management stores sensitive data in an encrypted form, ensuring that the actual values of the secrets are not exposed. This adds an extra layer of security when handling sensitive information like API keys or access credentials. With GitHub Codespaces, it's possible to define access policies and permissions for secrets and explicitly control which secrets are accessible to the Codespaces environment.

The REST API also provides a way to track and monitor the usage of secrets, allowing the Auto-GPT community to identify any unusual activity or potential security concerns. This can help you detect and address AI-escaping scenarios promptly.

[Pwuts]

Auto-GPT has Docker integration now and can be run in a Docker container or even devcontainer. :)

[Torantulino]

Interesting. Which code specifically are you talking about separating?
Do you mean to run AI generated code using Codespaces or to execute the entire program there?

[slavakurilyak]

@Torantulino I was thinking of separating code from docker files

[AndresCdo]

We can use Docker under strict security enhancements or remotely.

[slavakurilyak]

I strongly recommend GitHub Codespaces for superior isolation and security to prevent AI-escaping scenarios. It offers a fully managed, user-friendly solution with tight GitHub integration, while local Dockerfiles pose higher risks due to the potential exposure of sensitive data.

[Taytay]

@slavakurilyak : I think a codespace would indeed be helpful...however, doesn't that also have access to some of your sensitive Github env variables? I like the idea of a totally-sandboxed docker container for running all of this...

[Taytay]

I'm definitely a fan. I also think we should define a devcontainer to go along with it so that local development can happen in Docker...