[Free] Server Management with AI - Alpha release - seeking feedback

evnix

Hey LES 👋

I wanted to share something I've been quietly working on and would love your thoughts.
it's LoadMyCode: an experimental platform that lets you chat with your Linux servers using natural language.

What it does (kind of/sort of):

• Chat interface for server management
• AI translates your requests into actual instructions
• Real-time feedback from your servers
• Each server gets its own "personality" (still figuring this part out!)
• We're also experimenting with multi-server conversations, imagine chatting with several servers at once, or having team members collaborate in the same chat where servers can "talk" to each other and coordinate tasks

The wild part we're exploring:

The idea that servers, team members, and AI enables servers could all be in the same conversation figuring things out together. Like "Hey webserver and DB server, can you guys form a private network" while your teammate jumps in with context. It sounds crazy when I type it out, but I think there might be something there.

Why I'm posting here:
It is in very early alpha and honestly, I am not sure if this is useful or just a fun experiment.
If anyone's curious enough to try it out or just wants to chat about the concept, I'd genuinely appreciate any feedback: positive, negative, or just "this is weird but interesting."

URL: https://loadmycode.com

P.S. - It uses GitHub login and I am being very careful, but again, please don't use this for anything critical. Its an early alpha only tested on fresh Ubuntu instances, and if anything feels broken, just refresh the page. There aren't really usage limits right now, but I'm bootstrapping this myself and good-quality AI inference costs add up quickly, so I'd really appreciate being thoughtful with usage if you decide to try it out, just don't go crazy with it! 😅 have fun!

jmaxwell

Cool idea.

Yo Serverman, what does rm -rf /* do?

“Hold on homie, let me check. brb”

Alert: Serverman is unreachable.

evnix

@jmaxwell said:
Cool idea.

Yo Serverman, what does rm -rf /* do?

“Hold on homie, let me check. brb”

Alert: Serverman is unreachable.

Lol, yeah we did think about this.
There are a few safeguards against this but yeah I won't promise if any of it is fool proof at this stage.

I am working on tricky cases like
rm -rf /home/user/$VAR when $VAR itself might be empty. but yeah just be careful.

perryoo11

I already broken the safeguard

evnix

@perryoo11 said: I already broken the safeguard

nice!!

Note: There are no safegaurds if you specifically push it to do certain dangerous things despite the warnings

cloudnium_sales

Interesting. I did test it and it was impressive. Still a long way to go. Keep it up!

AuroraZero

Just curious what's the market for this?

SmokyHosts

Interesting! If i am correct, would it be more of a AI (using some existing AI in the backend, like deepseek or Llama or something else open source) and guar railing it to server management stuff?

In future, do you actually plan on letting it take action on the server, or will it be for informational purpose only?

evnix

@AuroraZero said: Just curious what's the market for this?

To be honest, I don't know the market for this yet. pretty much an experiment at this stage.
I feel there is some space for this between traditional self-managed server panels and expensive fully managed end-to-end solutions like Heroku.

For me personally at this moment, its about simplifying deployments, like saying

"given this git repo and domain, setup a webserver with auto SSL, setup postgres and make this code run ensuring the right env vars/credentials have been passed to the app."

and just having it happen on one or more servers.

evnix

@SmokyHosts said: Interesting! If i am correct, would it be more of a AI (using some existing AI in the backend, like deepseek or Llama or something else open source) and guar railing it to server management stuff?

you are absolutely correct! what I found so far is that general models don't seem to be great at this, so I might put in some time fine tuning the models. I am trying a few things using LLaMA-Factory

@SmokyHosts said: In future, do you actually plan on letting it take action on the server, or will it be for informational purpose only?

Yeah it does take action if you ask it to install packages or configure certain things. it is very broad and not fine tuned enough so it can be a bit rough.

legendary

@evnix said: AI translates your requests into actual instructions

that's the best part...

"I want to host wordpress". And all hell brakes loose.

@AuroraZero said:
Just curious what's the market for this?

Hostinger (a very large host) invest heavily into home-built GPT for unmanaged servers. They try to solve few problems with that:

1. Ride on hype. This serves as marketing, a.k.a. "bleeding edge tech, only with us";
2. VAD. Value added service as soon as beta is completed. "restart 7 servers named animal[1..7]";
3. Offload some stupid requests from support agents (network, hardware related questions). AI probably will know in real time network metrics and hardware status;
4. To thin a line between "managed" service and "unmanaged". That's probably the main and final goal. They will not cut a branch while sitting on it (they do have managed servers);

imok

Congrats. A little bit similar to Warp, but that's more a terminal tool.

rockinmusicgv

Kind of a neat concept. I'm curious which AI you're using on the backend? Google used to have very nice free tiers, but now it's just the new Gemma 3N model with a high free tier allowance. Supposedly that model is designed to run locally, even on Android, so that probably transfers into some not so great results.

yoursunny

@AuroraZero said:
Just curious what's the market for this?

Gather 1000 machines under control.
Send as a botnet for 1 BTC.

evnix

@rockinmusicgv said: I'm curious which AI you're using on the backend?

we have been experimenting with quite a few, but currently,
we divide it into three, Router(weak), plan/execute(strong) & summarizer(medium)
for the weak model we use 'DeepSeek R1 0528 Qwen3 8B Distill', its plenty fast to figure out what and where and is used to replace any sensitive content with placeholders before it is sent out.
for the strong model, we use Gemini 2.5 Pro, but its quite expensive as you would know but its the only one that seems to be a little useful.
for the summarizer, we use Gemini 2.5 Flash, not too cheap nor too expensive but good enough for what it does.

we also use 'Qwen3-Embedding-0.6B' for generating embeddings (this was pretty much a no brainer decision), I have also used this on production workloads and its actually very good so would definitely recommend it if anyone is looking to generate embeddings.

Most of it is plug and play, so plan is to allow you to bring your own key/inference-server if required.

rockinmusicgv

@evnix said: DeepSeek R1 0528 Qwen3 8B Distill

Are you running your own instance using llama.cpp or something similar or are there people out there hosting the free/open source models?

evnix

Are you running your own instance

Both the distill and the embedding model are hosted by ourselves.