I’m sharing a small golang experiment : MyCodeInterpreter. It connects ChatGPT with your own server, allowing for shell command execution on your machine.
- Has built-in ngrok for easy and secure “https” URLs behind firewalls and nat.
- Generates an OpenAPI schema for integrating with a custom GPT Bot.
- Runs easily in docker
This obviously comes with a big warning sign. Dont run it unless you read through the code understand the dangers
Open source projects always get a like from me!
lots of security challenges to consider even after hardening, but the use case and approach is a legit game changer for platform engineering, SOCs, cloud operations and the devops world. notionally the various “agent” providers in the observability and IT space e.g. Puppet, New Relic, DataDog, et al. are going to quickly realize the advantage of a conversational ui interface to their tooling.
Yeah I couldn’t really imagine what it would take to leave it on in ‘nosafe’ mode on a server. However it’s a very tempting thought having that chat in my phone at all times, preferably with full root access and a fair amount of SSH keys This thought forced me to try it out at least.
With static url to it , not having to paste the new openapi spec in after every launch , it might be a useful tool to start up once in a while though -to just have a convenient way to do a few tasks in ‘safemode’. So I’m probably going to work on adding support for another tunneling provider than ngrok since they don’t have any cheap premium options. It has to be simple and quick to launch, so having a built-in tunnel support is important I think for this usecase. Adding built-in ip restrictions would be next thing
The other use case would be just having it running in a locked down Docker with internet access , fairly safe and you could do more things than currently possible in the openai code interpreter. However I guess there are many projects that does that and I’m not sure the chatgpt interface is that interesting perhaps…so better to just use the API for that
I am really blown away by how easy it is to do a thing like this though with the product openAI has made. The new Gpt bot thing and it’s support for openapi actions are insainly powerful in its simplicity
Can you give more examples of using this versus just asking code questions to GPT? I’m not quite grokking it. It seems worth thinking about a bit more though.
I had a similar idea, but for local files. Asking ChatGPT to go through every code file in a project and add documentation is just
I’m still tuning it a bit, but I expect it’s about at a point where it can implement features semi-autonomously.
@mail.reknew - ChatGPT is executing commands on the server. No copy-pasting.
yeah documentation was bad , and still is. I have added a video now though that hopefully gives some intuition on how it works and what it does. ALso the docker solution works better now , this makes it safer to try ofc, but there are still no security on the ngrock entrypoints.
Thanks for the video, that’s great. And now I found sgpt because of you too, woohoo!