4 GB of RAM is all it takes to have a private AI chat running on your own machine. The d0a1 packs repository contains four Docker-ready deployments: openwebui, wordpress, hermes, and openclaw. Each pack installs with a single script, and the same command that spins it up on your computer today keeps it running 24/7 on a VPS tomorrow. This guide walks through that full workflow: Docker, cloning the repository, local deployment, and the jump to a VPS.
Prerequisites
Before opening the terminal, check three conditions. Memory is the real constraint: every pack starts with 4 GB of RAM, except WordPress, which requires 6 GB because it adds a database and an AI model to the same machine. Docker is mandatory in all four cases, and we install it in step 1. You only need an internet connection the first time, when Docker downloads the images; after that it reuses the ones already on disk.
Step 1: install Docker
On Ubuntu or Debian, the official installer gets Docker ready in a few minutes:
curl -fsSL https://get.docker.com | shVerify the installed version:
docker --versionIf you’ll be using the machine with a non-root user, add it to the docker group and log back in so the change takes effect:
sudo usermod -aG docker $USERStep 2: clone the d0a1 packs repository
The public repository is called d0a1 packs and is hosted at git.d0a1.es. Clone it and enter the folder:
git clone https://git.d0a1.es/d0a1/packs.git
cd packsEach pack lives in its own folder inside the repository, along with its corresponding deploy.sh script.
Step 3: pick a pack and deploy locally
Enter the folder of the pack you want to try and run the script in local mode. With openwebui, the lightest example:
cd openwebui
./deploy.sh localThe script downloads the images and spins up the containers; the initial download is the slow part. To try another pack, repeat the same two commands inside the wordpress, hermes, or openclaw folders.
What each pack installs
- openwebui: a private AI chat in the browser. It combines Ollama and Open WebUI, and asks for no account or API key because the model runs on your machine.
- wordpress: WordPress alongside MariaDB 11.8, plus Ollama with llama3.2:3b as the default model and Open WebUI with the AI Agent Bridge plugin, which connects the AI to the CMS to draft posts and moderate comments.
- hermes: the CLI for the Hermes agent from Nous Research. After deployment, setup follows three commands: hermes setup, hermes model, and hermes gateway setup.
- openclaw: the OpenClaw assistant connected to messaging apps: WhatsApp, Telegram, Signal, and Discord.
Step 4: move to a VPS for 24/7 hosting
Once the pack works on your machine, vps mode moves the deployment to a remote server, where the service stays up around the clock:
./deploy.sh vpsThe memory requirement doesn’t change: 4 GB minimum and 6 GB for WordPress. If you don’t have a VPS yet, Vultr and DigitalOcean are two popular providers; you can sign up using the affiliate links https://d0a1.es/go/vultr/ and https://d0a1.es/go/digitalocean/ and pick a machine with at least that much memory.
Common issues
Docker responds "permission denied": your user isn’t in the docker group. Add it and log back in:
sudo usermod -aG docker $USERThe script won’t start due to missing execute permissions:
chmod +x deploy.shThe port is already in use, something common when Apache is holding port 80 and you want to deploy wordpress. Stop the service blocking the port:
sudo systemctl stop apache2WordPress won’t start or keeps restarting: check available memory, because this pack needs 6 GB:
free -hThe first run takes a long time and that’s not an error: Docker downloads the images and Ollama pulls llama3.2:3b only at that point. Subsequent runs don’t repeat the download.
Conclusion
Start with openwebui today: it’s the lightest pack and it verifies that Docker works properly on your machine. When you want a permanent service, get a VPS with the right amount of memory and run ./deploy.sh vps inside the pack’s folder. The only difference between a local test and a permanent service is the argument you pass to the script.