Combined smart-home server hardware

Best Mini PC for Home Assistant, Frigate and Local AI

Running Home Assistant, Frigate and local AI on one mini PC is a consolidation problem, not a normal mini-PC shopping problem. Home Assistant itself is usually light, Frigate adds continuous video decoding plus detection, and Ollama or local voice AI can consume the largest block of memory or GPU capacity. This guide sizes the machine from all three workloads together, then surfaces complete live systems near the top so the recommendation is tied to hardware you can actually buy.

See live hardware Size my workload

Quick answer

What this combined workload needs

For Home Assistant plus a modest Frigate setup, 16GB can be enough; once local LLMs, semantic search, face recognition or several other containers are added, 32GB is the more comfortable starting point. A low-power Intel host is attractive for camera decoding, while Core Ultra, Ryzen AI or a GPU-capable mini PC makes more sense when local AI is a first-class workload.

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

Choose the architecture before the brand. Decide whether Frigate detection uses Intel OpenVINO, Hailo or a GPU, whether Ollama runs CPU-only or on a supported GPU, and whether Home Assistant OS lives in a VM or Home Assistant Container lives beside the other services. Then buy the smallest complete host that leaves real memory, storage and thermal headroom.

Interactive decision tool

Mini PC Stack 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.

Three applications create three different bottlenecks

Home Assistant automations and dashboards rarely determine the CPU tier by themselves. Frigate is sensitive to hardware video decoding, detector access and camera concurrency. Local LLMs care much more about model size, memory bandwidth, available RAM or VRAM, and context length. Treating all three as one generic “server workload” hides the reason a cheap mini PC succeeds for one person and feels unusable for another.

The practical sizing method is to budget each workload independently and then add headroom for overlap. A doorbell event can trigger Frigate, an automation, a notification and a local AI request at the same time. The machine needs to remain responsive during that overlap, not only while each service is idle.

N100 and N150 are appliance-class choices

Intel N100 and N150 mini PCs are attractive when Home Assistant is the main service and Frigate has a small number of cameras using hardware decoding plus a separate detector. Their low idle power makes them good 24/7 appliances, but they leave less CPU and memory bandwidth for larger local models or multiple heavy containers.

Use this tier for smart-home control, a few cameras and lightweight AI assistance rather than trying to turn it into an all-purpose inference workstation. If local LLM latency matters, spending more on compute can be more effective than over-optimizing the lowest possible idle wattage.

N305 and Core-class systems buy concurrency headroom

Moving to Intel N305, recent Core i5 or comparable higher-core hardware provides more room for Home Assistant apps, databases, Frigate enrichments and background jobs without changing the basic low-power architecture. The integrated graphics can still handle video decoding while a dedicated detector handles object inference.

This is often the best value tier when the goal is one quiet box for a family smart home rather than a local-AI workstation. It gives more margin for updates and temporary CPU spikes while keeping power and cooling easier than a discrete-GPU build.

Core Ultra and Ryzen AI are not interchangeable marketing labels

Core Ultra systems can expose Intel GPU and NPU resources that are useful to supported Frigate OpenVINO paths, while Ryzen AI systems may provide strong integrated graphics and, on some current platforms, Ollama-supported AMD GPU acceleration. The exact processor matters more than the word “AI” in the listing.

Do not assume an NPU accelerates Ollama just because the mini PC advertises one. Ollama’s official hardware support focuses on supported GPUs and APIs; Frigate has its own NPU/OpenVINO support. Match each accelerator to the software that can actually use it.

RAM should be sized for the LLM, not just Home Assistant

A Home Assistant-only machine can be comfortable with far less memory than a local AI server. Once an Ollama model stays resident, model weights, context, KV cache and other containers compete for RAM or VRAM. Home Assistant also defaults its Ollama integration to an 8k context, and larger contexts increase memory use.

For a combined server, 32GB is a practical general target when local AI is modest. Move toward 64GB when you plan larger CPU-resident models, many services, larger contexts or virtualization. If the model fits mostly in GPU memory, host RAM still needs enough room for Home Assistant, Frigate and the operating environment.

Frigate video decoding and object detection are separate

A Hailo accelerator can reduce object-detection load but it does not decode camera streams. Intel, AMD or NVIDIA media hardware still needs to be exposed correctly for ffmpeg decoding. Likewise, a powerful GPU can be wasted if Frigate is accidentally decoding in software.

Pick the media path and detector path separately. A good low-power design often uses an Intel iGPU for decoding and Hailo or OpenVINO for detection, preserving larger GPU resources for local AI only when they are actually needed.

Local AI can dominate the thermal design

Ollama inference can hold the CPU or GPU at sustained load for much longer than ordinary Home Assistant activity. Mini PCs that feel silent during automation workloads may ramp fans heavily during local model generation. A system with adequate heatsink area, airflow and replaceable fans can be a better server than the smallest chassis on the product page.

If you expect frequent LLM use, compare sustained behavior rather than short burst benchmarks. Heat from NVMe drives and an external accelerator also accumulates inside compact cases, especially when the machine sits in a network cabinet.

Radio devices need a clean USB or network strategy

Zigbee, Thread and Z-Wave radios introduce device-access requirements that do not show up in CPU benchmarks. Direct USB passthrough is simple on bare metal but becomes another layer of configuration in a VM. Network coordinators can reduce USB passthrough dependence in some designs.

Plan radio placement for signal quality as well as server architecture. A USB coordinator buried beside a noisy mini PC or metal rack may perform worse than one on a short extension or a network coordinator placed centrally. Hardware consolidation should not damage the wireless side of the smart home.

Storage should separate system data from bulk recordings

Home Assistant databases, container layers, Ollama model files and Frigate metadata benefit from reliable SSD storage. Frigate recordings can consume far more capacity and may belong on a separate surveillance HDD or dedicated storage volume. Mixing everything on one nearly full SSD makes capacity management and recovery harder.

A 1TB system SSD gives useful flexibility for a multi-service host; 2TB or 4TB can make sense when local AI models and short Frigate retention also live there. Long video retention should still be calculated separately rather than treated as “whatever space is left.”

2.5GbE is useful for the stack, not required by one camera

Individual cameras usually do not need multi-gigabit links. A 2.5GbE mini PC becomes useful when the same host backs up Home Assistant, serves Frigate recordings, pulls model files, accesses NAS storage and streams several clients at once.

Buy multi-gig networking because the combined server moves more data, not because Home Assistant needs it. If recordings remain local and the home has ordinary camera bitrates, reliable wired gigabit may still be sufficient.

Virtualization trades flexibility for device complexity

Home Assistant OS is the recommended Home Assistant installation for most users and works well as a VM. Frigate documents that VMs add overhead and device-passthrough complexity, although Proxmox QEMU deployments are common. Ollama GPU passthrough adds another device-ownership decision.

A dedicated Linux/Docker host can be simpler when Frigate and Ollama are the primary workloads. Proxmox makes more sense when snapshots, isolation and multiple VMs are valuable enough to justify deliberate iGPU/GPU/USB passthrough planning.

Buy for the next workload, not the current idle graph

Smart-home servers tend to accumulate services. Voice pipelines, more cameras, a larger local model or a new radio can turn spare capacity into a constraint quickly. Favor replaceable memory, a second storage slot, 2.5GbE and useful accelerator connectivity when the price difference is reasonable.

The goal is not to overbuy a gaming machine. It is to avoid a mini PC whose soldered RAM, single SSD slot or inaccessible accelerator path forces replacement as soon as the combined stack becomes more ambitious.

Questions people ask

Best Mini PC for HA, Frigate & Local AI questions

How much RAM do I need for Home Assistant, Frigate and Ollama?

32GB is a practical combined starting point for modest local AI. 16GB can work when Ollama use is very light; 64GB is more comfortable for larger CPU-resident models, virtualization or many services.

Is Intel N100 enough for Home Assistant and Frigate?

It can be for a small, well-accelerated Frigate setup plus Home Assistant. Add local LLMs and the CPU/memory headroom becomes much more limited.

Do I need a GPU for the combined stack?

Not always. Frigate can use Intel/OpenVINO or Hailo, and Ollama can run on CPU. A supported GPU matters when local model speed or heavy enrichments are priorities.

Does an NPU accelerate Ollama?

Do not assume that. Frigate can use supported Intel NPUs through OpenVINO, while Ollama documents its own supported GPU paths. Verify software support for the exact accelerator.

Should Home Assistant run in a VM?

Home Assistant OS works well in a VM and is the recommended installation type for most users. Device passthrough becomes the main design concern in a combined server.

Can Frigate and Home Assistant share one mini PC?

Yes. Frigate even documents same-device Docker Compose integration patterns. Size the host from cameras, detector, memory and storage rather than Home Assistant alone.

Where should Ollama models be stored?

On fast local SSD storage when possible. Model libraries can grow quickly, so leave capacity headroom or use a dedicated model-storage path.

Do I need 2.5GbE?

Not for basic Home Assistant or individual camera streams. It is useful for NAS recordings, backups, model transfers and a broader home-server role.

Should I use a Hailo accelerator and a GPU together?

That can be sensible: Hailo for Frigate detection and a supported GPU for Ollama or enrichments. The host must have the required interfaces and cooling.

What is the best upgrade from a Raspberry Pi?

A complete x86 mini PC with replaceable RAM/SSD and hardware video decoding is often a practical step when cameras or local AI outgrow the Pi.

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.