Run your stack at home
Every tool in our dataset runs somewhere. For a growing slice of the agent stack, the best place is a quiet box in your own house: n8n's self-hosted Community edition is free with unlimited executions, Home Assistant keeps your home data at home, locally hosted MCP servers cost nothing per call, and Ollama runs open-weight models with no API bill and no prompts leaving the building. This page specs the hardware for that — plus the two books we think are worth the shelf space above the desk.
It pairs with the Zapier vs Make vs n8n comparison, which does the cloud-versus-self-host cost math that leads here.
Sizing, briefly
RAM is the spec that matters. 16GB is the floor for an automation box: n8n, Home Assistant, and a handful of services fit comfortably and stay responsive. For local LLMs, memory is the model budget — go 32GB or more if you want 14B-class models with room to breathe. On Apple Silicon the memory is unified, shared between CPU and GPU, which is why the Mac's 24GB and 32GB configurations exist.
Storage: local models are big files and Docker images pile up; a fast 1–2TB NVMe drive is the difference between a stack you grow and one you ration. Power: an always-on box deserves a small pure-sine-wave UPS, so an outage becomes a graceful shutdown instead of a corrupted database.
The hardware
Beelink SER9 Pro
Local LLMs and heavy automation — the flagship home agent box
The most local-AI headroom in this list: 32GB of fast memory runs 14B-class models in Ollama comfortably, and the NPU takes always-on inference chores. It is also far more than enough for n8n, Home Assistant, and a stack of Docker services running at the same time.
$1,169.00
GMKtec NucBox K8 Plus
Self-hosted n8n and Docker stacks — the upgradable workhorse
Unlike the soldered-memory boxes, the K8 Plus takes standard SO-DIMM RAM and M.2 drives, so it can grow with your stack. Dual Ethernet and an OCuLink port make it the flexible pick for a home server that also runs your automation around the clock.
$919.99
Apple Mac mini (M6)
Quiet always-on agent host — local models with the least fuss
The current Mac mini — Apple replaced the M4 model with the M6 in 2026, and this is the one it sells now. Silent under load, sips power, and its unified memory is fast for local models via Ollama or LM Studio. If local LLMs are the main job, step up to 24GB or 32GB of memory; storage is not upgradable after purchase, so size it honestly.
$899.00
NVIDIA Jetson Orin Nano Super Developer Kit
Edge AI and vision agents — a CUDA appliance, not a general server
A small AI computer for agents that see: camera feeds, on-device vision-language models, and robotics-adjacent projects, all inside NVIDIA's CUDA/JetPack ecosystem. It can run small local LLMs, but 8GB is tight — buy it for edge inference, not as your main n8n box.
$249.00
Raspberry Pi 5 (8GB)
The cheap always-on node — Home Assistant, light n8n, MCP servers
The lowest-cost way to keep Home Assistant, a small n8n instance, and a few MCP servers running around the clock on a few watts. The price shown is Raspberry Pi's board-only list price — memory shortages pushed Pi prices up through 2026 — and you still need the 27W USB-C power supply, cooling, and storage. Most buyers grab a kit that bundles those, which is what the Amazon button searches for.
$175.00
Samsung 990 PRO 2TB NVMe SSD
Fast storage for a mini-PC build — OS, Docker images, and local model files
Local models are big files: a handful of 14B-32B models plus Docker images eats a 512GB drive fast. The 990 PRO is the safe, fast default for a mini-PC build or a Pi 5 NVMe HAT. Samsung's list price is shown; street prices usually run lower, which is what the Amazon button is for.
$639.99
CyberPower CP1000PFCLCD UPS
Clean backup power — ride out outages and shut the stack down gracefully
An always-on stack that loses power mid-write eventually corrupts something. This pure-sine-wave UPS keeps a mini PC, router, and switch alive through short outages, and signals the host over USB to shut down cleanly when the battery runs low.
$199.95
Two books worth the shelf space
AI Engineering — Chip Huyen
The production-AI reference
For builders moving from API demos to production AI applications: evaluation, RAG, agents, and inference trade-offs, from a former Stanford ML systems instructor.
AI Agents in Action, Second Edition — Micheal Lanham
The hands-on agent-building guide
For Python developers building agents with today's stack: tool use, memory, MCP, and multi-agent patterns, worked end to end. The second edition is the one to get — it adds the MCP and agent-to-agent coverage the 2025 first edition predates.