Home Assistant, Frigate & Ollama Server Build

Three-workload local server build

A Home Assistant + Frigate + Ollama server combines a lightweight automation platform, a real-time video pipeline and an on-demand model inference engine. The workloads peak differently and can compete for the same GPU, RAM and storage. A successful build assigns each accelerator a job and preserves enough headroom that an Ollama request does not disrupt camera detection or automations.

See live hardware Size my workload

Quick answer

What this combined workload needs

For a serious three-service build, 32GB RAM is the practical floor I would target, with 64GB appropriate for larger CPU-resident models or heavy virtualization. Keep Frigate video decoding on an iGPU or supported GPU, use a suitable detector for object inference, and reserve a supported GPU for Ollama when low response latency matters.

Live Amazon hardware

Current products that fit this decision

Compare current Amazon listings relevant to this guide. Product availability and prices can change.

Checking the dedicated Home Assistant + Frigate + AI catalogue…

Buying decision

Match the hardware to the bottleneck

The cleanest architecture often uses separate acceleration paths: iGPU for video decode, Hailo/OpenVINO for Frigate detection, and a supported NVIDIA/AMD GPU for Ollama. If the host cannot expose those devices cleanly or thermally sustain them, split Ollama onto a second machine instead of forcing everything into one chassis.

Interactive decision tool

Frigate + Ollama Resource Planner

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.

Ollama is an external service from Home Assistant’s point of view

The Home Assistant Ollama integration connects to an Ollama server by URL. That server may be on the same physical machine, another VM or a separate computer. This is useful because Home Assistant does not need to own the model runtime in order to use it as a conversation agent.

Treat Ollama as a service endpoint when designing the stack. That makes later migration easier: the Home Assistant configuration can stay largely intact while the model runtime moves to a faster GPU box.

The LLM can become the largest memory consumer

Model weights and context consume much more memory than ordinary Home Assistant automations. Ollama documents that larger context windows require more memory, and its current context defaults also depend on available VRAM.

Do not size host RAM from Home Assistant and Frigate alone. Decide the model class and expected context first, then reserve memory for the rest of the system. CPU-resident models make 64GB hosts attractive sooner than users expect.

GPU allocation is the central architecture decision

A discrete GPU can run Ollama and may also be usable for Frigate detection, decoding or enrichments, but sharing one device creates contention and a single driver dependency. A dedicated detector plus iGPU media path can leave the larger GPU focused on model inference.

This separation is especially useful when someone speaks to the local assistant while several cameras become active. Frigate latency stays predictable instead of competing with a long LLM generation.

Frigate’s detector does not need to be the Ollama GPU

Hailo and Intel OpenVINO are efficient object-detection paths. They can reduce the need to dedicate a large NVIDIA card to camera inference.

If the machine already contains a supported Intel NPU/iGPU combination, Frigate can use those resources while Ollama uses a separate supported GPU. Assigning work by specialization often improves both power efficiency and troubleshooting.

CPU still matters after you add accelerators

Home Assistant logic, Frigate orchestration, ffmpeg tasks that are not accelerated, databases, Docker, MQTT and model preparation still use CPU. A low-end processor can become the orchestration bottleneck even when inference runs elsewhere.

For a three-service server, N305 or Core/Ryzen-class hardware provides more comfortable background headroom than the smallest appliance CPU. The exact tier depends on cameras, model usage and other services.

SSD capacity must include model files and container growth

Ollama model libraries can consume tens or hundreds of gigabytes as users experiment. Frigate and Home Assistant also create databases, metadata, snapshots and container layers. A 512GB SSD can feel cramped quickly on an all-in-one host.

1TB is a practical baseline; 2TB or 4TB gives more freedom for models and short recording retention. Keep long Frigate footage on storage sized from bitrate and days rather than letting it compete with model storage.

Home Assistant entity exposure should stay deliberate

Home Assistant’s Ollama integration can expose entities for local model control, but current guidance recommends limiting exposed entities for experimental control and notes that smaller models make more mistakes.

Treat the LLM as a tool-enabled integration rather than an omniscient administrator. Expose only what the model needs and keep safety-critical automations deterministic when possible.

Voice pipelines and LLM chat are not the same workload

A fully local Assist pipeline may use speech-to-text and text-to-speech without requiring a large conversational model. Ollama adds open-ended language understanding and tool use.

If the goal is fast local voice control, optimize the speech pipeline and Home Assistant first. Add a larger LLM only for the conversational tasks that justify its latency, memory and power cost.

Docker makes same-host deployment understandable

Frigate and Ollama both document Docker paths, and Home Assistant Container can also live on a Docker host. This creates a coherent Linux architecture for advanced users who are comfortable managing their own updates and device mappings.

The tradeoff is losing the Home Assistant OS apps experience and taking responsibility for the host. A VM-based HA OS design can preserve that experience while Frigate/Ollama remain separate services.

Virtualization magnifies GPU ownership questions

A GPU generally cannot be casually shared across several VMs without specific platform support. Passing it to Ollama may prevent Frigate from using the same device, and vice versa. iGPU and USB/Hailo passthrough add more mappings.

Design the device ownership table before creating VMs. “I will pass the GPU through later” is not a deployment plan when three services may want it.

Power and noise change once local inference is frequent

A low-power surveillance host may idle quietly until Ollama begins generating tokens. Sustained GPU or CPU inference raises wall power and fan speed, which affects UPS sizing and where the machine can physically live.

Estimate annual power from average use rather than peak TDP. If local AI is queried only a few times per day, a higher-power GPU may still have a modest energy impact; if it runs agents constantly, efficiency matters much more.

A split Ollama server is often the clean upgrade path

If the smart-home/NVR machine is stable and only local AI is outgrowing it, move Ollama to a second GPU system. Home Assistant can continue connecting over the LAN, and Frigate does not have to inherit the new GPU driver stack.

Consolidation is valuable when it reduces hardware and idle power. It stops being valuable when every upgrade to the AI side risks the automation and camera side.

Questions people ask

Home Assistant, Frigate & Ollama Server questions

Can Home Assistant connect to Ollama on another PC?

Yes. The integration connects to an external Ollama server URL, so the model runtime can be on another local machine.

How much RAM for Home Assistant, Frigate and Ollama?

32GB is a practical starting point; 64GB is safer for larger CPU-resident models, many services or virtualization.

Can Frigate and Ollama share one NVIDIA GPU?

They can in some architectures, but contention and device ownership matter. A dedicated Frigate detector can leave the GPU free for Ollama.

Does Hailo help Ollama?

No. Hailo is useful for supported Frigate detection. Ollama uses its own supported CPU/GPU acceleration paths.

Should the Ollama models live on NVMe?

Yes when possible. Fast local SSD storage improves model loading and keeps the model library separate from long-term Frigate footage.

Is 512GB enough?

It can be for a small model set with recordings elsewhere, but 1TB or more is easier for a multi-service server.

Can I run all three in Docker?

Advanced users can run Home Assistant Container, Frigate and Ollama on Linux/Docker. Home Assistant OS remains the recommended Home Assistant installation for most users.

Can Home Assistant control devices through Ollama?

The Ollama integration can expose Home Assistant entities to supported tool-using models, but control is experimental and entity exposure should be limited.

Do I need a large LLM for local voice control?

No. Home Assistant can run a fully local voice pipeline without a large conversational model. Use Ollama when you want open-ended language capabilities.

When should I move Ollama to another machine?

When model size, GPU requirements, power or driver maintenance begin to compromise the stability of the Home Assistant/Frigate host.

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.