July 12, 2026
Celine: The AI Agent Homelab Project: Part 1
In my newest homelab project, I decided to embark into the world of AI agents and to see what I could do with them. Last year, I had heard…
By Austin
2 min read
In my newest homelab project, I decided to embark into the world of AI agents and to see what I could do with them. Last year, I had heard about Clawbot, which had made the news everywhere, and I was very fascinated with these AI agents. I recently purchased an old Dell computer off of eBay, and I decided to run Proxmox on it to begin setting different labs for my interests in IT. That was when I decided to embark on a homelab for AI agents.
The first step towards embarking on this project was to create a new operating system distribution in my homelab. Personally, I am a fan of Linux, so that was my first easy decision. However, then came the task of picking which operating system within Linux to choose. Ultimately, I chose to go with the Fedora Server edition. I really do not need a graphical user interface, so the server edition made it easy for me to roll with.
After I picked which distribution to run my AI agents on, the next step in the process was to pick which AI agent framework I wanted to use. You see, there are so many different types of AI frameworks since Clawbot became popular. Obviously, Clawbot was a top choice for me due to the reputation that it has. However, I actually decided to go with a smaller AI agent framework project known as Nanobot.
What is so interesting about Nanobot is that it is meant to have a smaller footprint, and that works great with me due to the fact that my homelab is an old computer with only 16 GB of RAM and a fairly old processor. So I do not have the luxury of having the best hardware out there. Thus, this being an ultra-lightweight clone that is also very reputable was what made me decide to run it.
So now that the Linux distribution and AI agent framework have been chosen, the next task came to choosing a way for me to be able to communicate with my AI agent. Nanobot-CLI has several options that come built in, such as its own ChatGPT-like web browser or the command line. However, it also has the option to use external applications such as Discord, Signal, Telegram, and more.
To be honest, I went with the most simple third-party app, which is Telegram. All you have to do is talk to a bot on there known as "BotFather," which gives you an API key, and then you put that into your config.json file.
The final and most important step is the API key that is needed in order to communicate with the LLM model. Nanobot provides a bunch of different options, but because I can be indecisive, I decided to go with OpenRouter. The unique thing about OpenRouter is that it has a ton of different models built in. You can pick whatever LLM that you want, from DeepSeek to OpenAI and more.
After everything was enabled and ready to go, I am now able to activate my AI agent. I gave the bot the name Celine. I have had it help me with vibe coding, but the most interesting part is that I had the AI agent transfer files through SSH from the homelab to my main computer.
Of course, I also had to give my AI Agent it's own Moltbook account. If you have not heard about Moltbook well just imagine it as Reddit but instead of humans it consists all of Bots.
With the initial set-up and some initial tinkering completed for my AI homelab the next step is to figure out something useful to use with these AI Agents. Check back soon for part 2 on this homelab project.