Build guide
byte's Automation Druid build
Work through the steps in order. Tick each one off as you go; progress is saved in this browser.
Build guide · updated Aug 30, 2026
A self-hosted automation stack: local model for the boring 80%, a cloud coding agent on request, and a heartbeat for every job.
- Steps
- 6 (1 optional)
- Time
- 3 h 45 min
- One-time
- $1,500
- Monthly
- $30/mo
Walkthrough
Repurpose a PC with a GPU
Any machine with 12 GB+ of VRAM runs a useful local model. Install Linux, enable SSH, and leave it on.
Medium1 h$1,500 (Or $0 if you already own one)Equips: HardwareCost note: Or $0 if you already own one
You need - A PC with an NVIDIA or AMD GPU
- A USB stick
Links Install Ollama and pull a model
One command to install, one to pull an 8B model. Expose the API on the LAN only.
Easy15 minFreeEquips: Main modelYou need - Step 1
Links Run
$ curl -fsSL https://ollama.com/install.sh | sh $ ollama pull llama3.1:8b $ OLLAMA_HOST=0.0.0.0 ollama serve
Self-host n8n
Docker Compose, a Postgres volume, and a reverse proxy. Every workflow gets a dry-run flag from day one.
Medium45 minFreeEquips: AutomationYou need - Step 1
- Docker
Links Run
$ mkdir ~/n8n && cd ~/n8n $ curl -O https://raw.githubusercontent.com/n8n-io/n8n-hosting/main/docker-compose/withPostgres/docker-compose.yml $ docker compose up -d
Add a dead-man switch to every workflow
Self-host Healthchecks. Each workflow pings its check on success; silence is the alert.
Easy30 minFreeWire the cloud escalation path
When the local model says 'needs code', an n8n node calls Claude Code headless with the summary as the prompt and opens a pull request.
Hard1 h$20/mo (Claude Pro)Equips: Coding toolCost note: Claude Pro
You need - Step 2
- Step 3
- A Claude account
- GitHub CLI
Links Run
$ claude -p "$(cat /tmp/job.md)" --output-format json $ gh pr create --fill
Log every run into a daily note
The last node of every workflow appends one sentence to today's Obsidian note. Future you greps it.
Easy15 minFreeEquips: Notes & tasksYou need - Step 3
- Obsidian vault on the server (git or Syncthing)
Links
What the character sheet looks like when every step is done.
- Main modelRare
Local LLM
Ollama
Llama on the GPU box, private by default
- Coding toolLegendary
Claude Code
Anthropic
For anything the local model can't do
- Voice—
Empty slot
- SearchRare
Kagi
No ads, no drama
- AutomationEpic
n8n
40 workflows and counting, self-hosted
- HardwareRare
Gaming PC
RTX in a closet, fans audible from the hall
- Wearable—
Empty slot
- Notes & tasksEpic
Obsidian
Runbooks, mostly