Home Assistant + local LLM mini PCs

Best Mini PC for Home Assistant and Ollama

Home Assistant needs relatively little compute; Ollama can need a lot. That makes this search less about “a fast Home Assistant box” and more about choosing a local model platform that still runs Home Assistant reliably. The right mini PC depends on whether the model runs on CPU, integrated graphics or a supported discrete GPU, how much context you want, and whether the machine also handles local voice services.

See live hardware Size my workload

Quick answer

What this combined workload needs

For lightweight local models and normal automations, a 32GB Core/Ryzen mini PC is a sensible starting point. If you want larger models or longer contexts without a discrete GPU, prioritize 64GB+ memory and memory bandwidth. If response speed matters, choose a mini PC or compact system with a GPU path Ollama officially supports.

Live Amazon hardware

Current products that fit this decision

These listings come from this sprint's dedicated Amazon catalogue. Barebones PCs, laptops, GPU enclosures, adapter-only accelerator listings, unrelated accessories and ambiguous radio products are excluded by the normalizer.

Checking the dedicated Home Assistant + Frigate + AI catalogue…

Buying decision

Match the hardware to the bottleneck

Do not pay extra for an “AI NPU” unless the software you plan to run can use it. Home Assistant benefits from the Ollama service, but Ollama’s documented acceleration paths center on supported GPUs. Buy from the model runtime backward: model class, context, acceleration, memory, then chassis.

Interactive decision tool

Home Assistant + Ollama Mini PC Selector

Size the combined workload before comparing products. Results are planning tiers, not benchmark guarantees. Verify current Home Assistant installation support, Frigate detector/device access and Ollama hardware support for the exact system.

Compatibility checklist

Four checks before you purchase

Assign every accelerator a job

Video decoding, Frigate detection, enrichments and Ollama inference are separate workloads. Do not assume one NPU/GPU automatically covers all of them.

Protect Home Assistant headroom

Size the combined host for peak overlap so camera or LLM activity cannot starve automations, databases or radio services.

Verify device access

USB radios, iGPU/NPU/GPU and Hailo devices must be visible to the correct host, container or VM before their benchmark performance matters.

Keep data roles separate

Use fast SSD space for system/models and size Frigate recordings independently so retention cannot crowd the smart-home environment.

Home Assistant is not the sizing problem

Automations, integrations and dashboards use modest resources compared with model inference. A mini PC that is overkill for Home Assistant may still be underpowered for the Ollama model you want.

Set a reliability reserve for Home Assistant, then size the rest of the machine for the model runtime. That keeps the smart home responsive even while a long response is being generated.

The Ollama integration expects an external server

Home Assistant connects to Ollama over a URL. The Ollama process can run on the same host or elsewhere on the network.

This gives you an important escape hatch: start with one box, then migrate Ollama to a dedicated GPU server without moving Home Assistant when local AI demands grow.

Model size and context are the first memory inputs

Larger models carry larger weights, and longer context requires additional memory. Ollama explicitly documents that increasing context length increases memory use.

Choose the model class you actually plan to keep loaded. Buying 64GB for a 3B model you rarely use may be unnecessary; buying 16GB for a larger CPU-resident model can turn every request into paging and frustration.

CPU-only inference is valid but changes expectations

Ollama runs without a GPU, so a high-memory mini PC can serve local models on CPU. That path is simple and can be efficient for occasional use, but token generation may be much slower than on supported GPU hardware.

CPU-only can be perfect for background summaries or infrequent requests. It is less satisfying when the local assistant is expected to answer quickly in a conversational voice workflow.

Supported GPU acceleration is the performance lever

Ollama currently documents supported NVIDIA, AMD, Apple Metal and experimental Vulkan paths. On a Linux mini PC, integrated AMD graphics or an attached/discrete GPU may be usable depending on the exact hardware and driver support.

Check the current Ollama hardware support list before assuming a particular iGPU works. Marketing phrases such as “AI engine” are not enough to establish acceleration.

Core Ultra NPU is not a universal Ollama shortcut

An Intel NPU can be valuable for applications that specifically support it, including some Frigate OpenVINO configurations. That does not automatically make it an Ollama accelerator.

If your main goal is local LLM speed, prioritize the compute path Ollama documents. A Core Ultra mini PC can still be a good host because of CPU, memory, media and efficiency improvements, but buy it for verified capabilities.

RAM capacity and memory bandwidth both matter

A model can fit in RAM and still generate slowly if the CPU and memory subsystem cannot feed it efficiently. Higher-capacity dual-channel memory and newer platforms can improve the experience even when the model size does not change.

For a service that also runs Home Assistant, leave several gigabytes free rather than sizing the model to consume almost every byte of host memory.

SSD sizing should account for experimentation

Users rarely download only one model. Different quantizations, embedding models, speech models and backups can turn a small model library into hundreds of gigabytes.

A 1TB NVMe drive is a comfortable general baseline. 2TB or 4TB is rational when the box is explicitly a local-AI server rather than just a Home Assistant appliance.

Local voice may need a different optimization

Speech-to-text and text-to-speech have different latency and hardware needs from an LLM conversation agent. Home Assistant can run a fully local voice pipeline without Ollama.

If fast device control is the priority, keep deterministic Assist commands fast and use the LLM for queries that benefit from reasoning. That can reduce the model size and hardware needed for a pleasant voice experience.

Entity exposure affects local model reliability

Home Assistant recommends exposing fewer than 25 entities when experimenting with Ollama control, and notes that smaller models make more mistakes.

Hardware cannot fix a tool-selection problem caused by giving a small model hundreds of similar entities. Good entity naming and limited exposure can improve the experience without buying a larger GPU.

Network and security still deserve attention

The Ollama API should be available to Home Assistant but not carelessly exposed to the public internet. Keep the service on trusted local networks and use sensible firewall rules.

If Ollama moves to another device later, a stable local DNS name is more maintainable than embedding changing IP addresses throughout integrations.

Choose a mini PC with an exit strategy

A mini PC can be the perfect 32GB starting point and still become an undersized AI host two years later. Replaceable memory, larger SSD support, USB4/OCuLink or a clean network path to a future GPU server make that transition easier.

The best purchase is not the machine that promises to run every future model. It is the one that runs today’s model well and lets Home Assistant remain stable when the AI side evolves.

Questions people ask

Best Mini PC for Home Assistant and Ollama questions

How much RAM for Home Assistant and Ollama?

32GB is a good general starting point for modest local models. 64GB+ is useful for larger CPU-resident models or longer contexts.

Does Home Assistant run Ollama directly?

The Home Assistant integration connects to an external Ollama server by URL. That server can be on the same physical host or another device.

Can I use Ollama without a GPU?

Yes. CPU-only works, but response speed depends heavily on model size, CPU and memory bandwidth.

Is an Intel NPU useful for Ollama?

Do not assume it is. Verify the acceleration path Ollama supports for the exact hardware and operating system.

Is Ryzen AI good for Ollama?

Some current Ryzen AI platforms have GPUs listed in Ollama hardware support. Verify the exact processor/GPU and driver path rather than buying from the Ryzen AI label alone.

How large should the SSD be?

1TB is a comfortable baseline; 2TB or 4TB gives more room for multiple models and backups.

Can I run Home Assistant OS and Ollama on one mini PC?

Yes through virtualization or by running Ollama elsewhere on the host/network. Plan device ownership and updates carefully.

How many entities should I expose to Ollama?

Home Assistant currently recommends exposing fewer than 25 entities when experimenting with local LLM control.

Do I need Ollama for local voice?

No. Home Assistant can run fully local speech-to-text and text-to-speech without an LLM.

When should I use a separate Ollama server?

When GPU, RAM, power or driver requirements exceed what you want on the always-on Home Assistant machine.

Official references and methodology

Verify current Home Assistant, Frigate and Ollama support

Cloudzat treats Home Assistant, Frigate and local AI as separate workloads that share one hardware envelope. Home Assistant installation guidance, Frigate device-access and detector guidance, and Ollama GPU/context documentation are used as the technical baseline. Calculator outputs are planning tiers rather than benchmark guarantees, and live Amazon listings come only from this sprint's dedicated catalogue.

As an Amazon Associate, Cloudzat may earn from qualifying purchases. Prices, seller terms, exact configurations and Home Assistant, Frigate or Ollama hardware support can change.