iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Most reliable installations use two roles: an always-on Home Assistant host and a separate computer that runs the local language model. Home Assistant Green, a Raspberry Pi, mini PC, NUC, or virtual machine can handle the automation service, while the model server needs memory sized for the model, quantization, context window, and number of simultaneous requests. A supported GPU can make replies faster, but a dedicated GPU is not mandatory in every configuration.
The hardware decision in one sentence
Choose the Home Assistant computer and the LLM computer separately unless one machine clearly has enough capacity for both. Home Assistant’s Ollama integration connects to an external Ollama server over your network, so the model does not have to run on the same device as Home Assistant. This separation lets you keep automations available on a low-power host while upgrading model hardware independently.
The correct specification depends on the exact model and quantization, the context length you intend to use, how many requests may run at once, and the response speed you consider acceptable. There is no universal minimum computer that covers every local model and Home Assistant combination.
What the Home Assistant host needs
Recommended host types
Home Assistant recommends Home Assistant Green for most people. It also supports installation on a Raspberry Pi, mini PC, NUC, or virtual machine. These recommendations describe the automation host, not a guaranteed local-LLM machine.
#1 Best Overall
- EVOLUTION RYZEN AI MAX+ 395 MINI PC - GMKtec EVO-X2 is the next evolution in AI mini PC Ryzen Strix Halo series. Thanks to AMD Simultaneous Multithreading (SMT) the core-count is effectively doubled, to 32 threads. Ryzen AI Max+ 395 has 64 MB of L3 cache and can boost up to 5.1 GHz, depending on the workload. The Ryzen AI Max+ 395 is currently rated as the "most powerful x86 APU" on the market for AI computing.
- AI NPU with XDNA 2 ARCHITECTURE - Powered by 16 “Zen 5” CPU cores, 50+ peak AI TOPS XDNA 2 NPU and a truly massive integrated GPU driven by 40 AMD RDNA 3.5 CUs, the Ryzen AI MAX+ 395 is a transformative upgrade and delivers a significant performance boost over the competition. The Ryzen AI Max+ 395 excels in consumer AI workloads like the llama.cpp-powered application: LM Studio. Shaping up to be the must-have app for client LLM workloads, LM Studio allows users to locally run the latest language model without any technical knowledge required and unleash their creativity and productivity.
- AMD RADEON 8090S iGPU GAMING PC - The AMD Radeon RX 8060S offers all 40 CUs with up to 2.9 GHz graphics clock and uses the new RDNA 3.5 architecture. The powerful iGPU is positioned between an RTX 4060 and 4070 laptop GPU and therefore enables gaming in FHD at maximum details in most demanding games. The 8060S can also utilize the full 128GB pool, which is perfect for running LLMs such as Deepseek 70B Q8, which runs comfortably on this machine.
- EIGHT CHANNEL LPDDR5X - LPDDR5X is a new ground breaking memory small form factor installed on-board. With blazing speeds up to to 8000MT/s, it runs 1.5x faster than the DDR5 SODIMMs; 90% better performance over DDR5 SODIMMs in video conferencing and photo editing; 30% better performance in productivity apps; 12% better performance in digital content workloads.
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-X2 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and dual USB 4 40Gbps Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
- Home Assistant Green: a purpose-built, always-on automation host. Treat it as the Home Assistant computer; the cited guidance does not establish it as an LLM inference system.
- Raspberry Pi: the Pi installation path asks for a Raspberry Pi 4 or 5 with at least 2 GB of RAM. That is a Home Assistant requirement and should not be read as enough memory for model weights or a large context.
- Mini PC or NUC: useful when you want more storage, memory, or virtualization headroom for Home Assistant and related services.
- Virtual machine: suitable when an existing computer already runs a hypervisor and can remain powered on.
Network and availability requirements
The Home Assistant host must be able to reach the computer running Ollama. Give the model server a stable address or a dependable local DNS name, and keep its service running whenever voice or dashboard requests should work. If the computer sleeps, changes address, or is blocked by a firewall, Home Assistant will not be able to obtain a response even when the model itself is installed correctly.
Use a wired connection where practical for a permanently running server. Wi-Fi can work, but long prompts and generated responses make an unreliable wireless link more noticeable. Network reachability is an integration requirement, not a model-performance upgrade.
What the LLM host needs
Memory is the first specification to size
The model computer needs enough available memory for the selected model and its runtime context, with additional headroom for the operating system and any other services. Model download size is not the same as the total memory requirement: the runtime may need space for loaded weights, the context cache, temporary buffers, and concurrent requests.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchOllama’s current quickstart gives Gemma 4 E2B as one concrete example. Its download is about 7.2 GB, and Ollama recommends 8 GB of available VRAM or Apple unified memory for that example. This is an example recommendation, not a floor that applies to every model. A different parameter count, quantization, or context setting can require substantially more or less memory.
Account for context length
Context contains the current conversation, system instructions, entity descriptions, tool definitions, and any retrieved history. A smart-home prompt can become surprisingly large when many entities and state details are exposed. Longer context consumes more memory even when the model weights do not change.
Ollama documents these context defaults by available VRAM:
| Available VRAM | Documented default context | How to interpret it |
|---|---|---|
| Below 24 GiB | 4k tokens | Lower memory use; long entity lists or histories may be truncated sooner. |
| 24–48 GiB | 32k tokens | More room for conversations and tool descriptions, with higher memory use. |
| 48 GiB or more | 256k tokens | Very large context is possible in the runtime, but the model may not use it effectively and memory demand is high. |
Ollama recommends at least 64,000 tokens for tasks such as agents and coding tools. That is a runtime recommendation, not a guarantee that every model, tool workflow, or Home Assistant conversation benefits from that window. Start with the smallest context that contains the instructions and entity information you actually need; increase it only after checking memory use and latency.
Free tools Windows power users keep installed
One-click scans. No signup required.
Plan for concurrent requests
If Home Assistant, a voice assistant, and another application can ask the model at the same time, size for concurrency rather than one prompt. Ollama’s FAQ states that required RAM scales with parallel requests multiplied by context length. Two simultaneous 16k-token conversations can therefore consume far more memory than one 16k request, even when the model weights are identical.
Rank #2
- EVOLUTION CORE ULTRA 9 285H MINI PC - GMKtec EVO-T1 is the next evolution in AI mini PC Ultra 9 series. The Core Ultra 9 285H offers 16 cores (six P-cores + eight E-cores + two LPE-cores) and 16 threads with a turbo clock of 5.4 GHz. It is currently one of the best value for performance AI mini PC computers.
- AI NPU - The 285H features an Intel AI Boost NPU, capable of up to 13 TOPS (Tera Operations per Second) for INT8 calculations, which is designed to accelerate AI tasks.
- INTEL ARC 140T GAMING PC - The Arc 140T GPU includes 8 Xe cores and supports features like DirectX 12, OpenGL 4.5, and OpenCL 3, making it capable of handling modern games and creative applications. It also supports Quick Sync Video for efficient video encoding and decoding, as well as AV1 encoding and decoding.
- 64GB DDR5 RAM + 1TB SSD - The EVO-T1 is equipped with Dual 32GB (Total 64GB) SO-DIMM DDR5 5600MHz memory sticks. 2TB PCIE 4.0 SSD Drive with 3x M.2 2280 Expansion slots. Each slot capable of reading up to 4TB. (12TB MAX)
- QUAD SCREEN 8K DISPLAY SUPPORT - EVO-T1 AI Mini PC support 4-screen 4K/8K output via HDMI 2.1 (8K@60Hz), DisplayPort 1.4 (4K@60Hz), and USB Type-C Transfer speed (supporting PD3.0/DP1.4/DATA). Ideal for gaming, video editing, and multitasking, it provides expansive and crisp multi-display support.
- Estimate the largest normal prompt, including entity descriptions and tool definitions.
- Multiply that context allowance by the number of requests that may overlap.
- Reserve memory for the operating system, Home Assistant add-ons or containers, and monitoring.
- Leave margin rather than sizing to an exact download number.
Do you need a GPU to run Ollama?
No. Ollama can use system RAM when available VRAM is insufficient, although it warns that responses may be slower. CPU or system-memory inference can be a sensible way to reuse an existing computer, especially with a smaller quantized model and modest expectations. It is less suitable when you need rapid voice responses, a long context, or several users at once.
When a GPU is worth prioritizing
A supported GPU reduces dependence on slower system-memory transfers and generally makes interactive inference more responsive. Compare the card’s usable memory and supported software path with the model’s requirements rather than choosing by brand name or price alone.
Ollama documents NVIDIA acceleration subject to the required compute capability and drivers. Apple devices can use Metal, and Vulkan support extends acceleration to additional Windows and Linux hardware. Confirm that the operating system, driver, and Ollama build support the path you intend to use before buying hardware.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why benchmark numbers need qualification
Ollama reported a configuration-specific scheduling result with Gemma 3 12B at 128k context on one NVIDIA GeForce RTX 4090: throughput changed from 52.02 to 85.54 tokens per second, while reported VRAM use changed from 19.9 GiB to 21.4 GiB. The figures come from that particular model, context, GPU, and software change. They are evidence that scheduling and context affect speed and memory, not a general performance promise or a recommendation that every buyer needs that card.
How Home Assistant connects to a local model
Ollama integration
Home Assistant’s Ollama integration adds a conversation agent that uses an external Ollama server. During configuration, point Home Assistant to the server’s reachable address and select the model installed there. The integration can use the Assist API and can be given access to information about, or control over, exposed entities.
Control is experimental. Only models that support tools can control Home Assistant, and smaller models are more likely to make mistakes. Home Assistant recommends exposing fewer than 25 entities while experimenting. Limit the initial set to devices whose incorrect operation is low risk, such as lights, and expand only after reviewing actual behavior.
llama.cpp and other OpenAI-compatible servers
Home Assistant also supports an external server that exposes an OpenAI-compatible chat-completions API. Its documentation names llama.cpp, llama-cpp-python, and vLLM as examples of compatible backends. If you choose Ollama, Home Assistant says to use its official Ollama integration rather than treating Ollama as a generic endpoint.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteAn OpenAI-compatible API describes how requests are sent; it does not ensure that a model can reliably call Home Assistant tools. Check tool-use support, function-calling behavior, authentication, and latency in the specific server and model combination you deploy.
Rank #3
- 【Low Power for Always-On AI Workflows】At just 15W TDP, the GEEKOM A7 uses far less power than a traditional 350W desktop, helping reduce electricity costs, heat, and cooling noise during extended operation. That efficiency makes it ideal for keeping cloud AI assistants and AI Agent tasks running in the background—automating document summaries, email polishing, meeting notes, content rewriting, research, and scheduled workflows throughout the day. The energy savings can help recoup the device cost in about 1 year, making A7 a practical choice for 24/7 AI task hosting and efficient everyday computing.
- 【Ryzen 7 7730U – More Than a Low-Power PC】Think low power means less performance? Not here. The Ryzen 7 7730U mini computer packs 8 cores, 16 threads, and up to 4.5GHz, giving you the power to handle multitasking, dozens of tabs, video calls, and creative work smoothly. AMD Radeon Graphics supports 4K playback, multi-display work, photo editing, and casual gaming without a dedicated GPU. Compared with the Ryzen 7 5825U and Ryzen 5 7430U, it delivers up to 20% higher performance for faster response and smoother everyday computing—all in a compact, energy-efficient Mini desktop.
- 【Lock In More Memory Before It Costs More】32GB gives you the headroom most demanding tasks need today—and room to grow tomorrow. Built for heavy multitasking, content creation, large projects, and AI-assisted workloads, the GEEKOM mini pc starts you with twice the memory of a typical 16GB setup, so you can skip an immediate upgrade. With AI driving greater demand for memory, starting with 32GB is a smarter way to stay ready for what’s next. The 500GB PCIe Gen4 x4 SSD delivers fast storage, with support for up to 64GB RAM and 4TB SSD storage when you need more.
- 【Premium Metal Design & 3-Year Warranty】Why settle for plastic? The GEEKOM mini desktop features a premium aluminum alloy chassis that resists daily wear and helps dissipate heat during extended use. Rigorous quality testing and CE, FCC, and RoHS compliance support dependable performance, backed by a 3-year limited warranty and professional support for long-term peace of mind.
- 【One Mini PC, All Your Ports】Stay connected with dual USB-C ports, 5 USB 3.2 ports, dual HDMI 2.0, and a 2.5G LAN port for fast, flexible connectivity. The USB-C ports support high-speed data transfer, display output, and peripheral power, while Wi-Fi 6E keeps streaming, file transfers, and online work fast and reliable. From multiple peripherals to high-resolution displays, everything you need stays within easy reach.
Four practical hardware layouts
| Layout | Best fit | Main trade-offs |
|---|---|---|
| Separate Home Assistant host plus GPU-capable LLM computer | You want an always-on automation appliance and faster local inference. | Higher purchase cost and power use, but clear role separation and independent upgrades. |
| Existing computer as the LLM server plus separate Home Assistant host | You already own a computer with suitable memory or a GPU. | Little added hardware cost; replies stop or slow when the computer sleeps, is busy, or is unavailable. |
| One adequately resourced computer for both services | You prefer fewer devices and can manage containers or virtualization. | Model inference and automation share CPU, memory, storage, and power. No cited source certifies a combined minimum specification. |
| CPU or system-memory inference | You prioritize reusing hardware and accept smaller models or slower replies. | It can work when VRAM is limited, but Ollama warns that system-memory fallback may reduce response speed. |
Evaluate each layout against six questions: Does the model fit in available memory at the chosen context? Can the server stay on? Is the network path dependable? How many requests can overlap? Is the response speed adequate for voice control? Would isolating Home Assistant reduce the impact of a model crash or resource spike?
A step-by-step sizing and installation plan
- Choose the model first. Record its parameter size, quantization, supported tool or function-calling behavior, and vendor memory guidance. Do not start with a computer specification and assume every model will fit.
- Define the Home Assistant workload. List the entities you expect the assistant to read or control, the maximum conversation history, and whether voice requests may overlap with automations or another user.
- Set a context target. Include system instructions, entity descriptions, tool schemas, and history. Use Ollama’s documented 4k, 32k, or 256k defaults only as starting points tied to the available VRAM bands; they are not universal model limits.
- Calculate concurrency headroom. Multiply the context allowance by the number of simultaneous requests, then add the model’s loaded-weight requirement and operating-system overhead.
- Select the compute path. Use a supported NVIDIA, Apple Metal, or Vulkan path when responsive GPU inference matters. Otherwise, verify that CPU or system-memory inference is acceptable for your expected response time.
- Install Home Assistant separately or on the same host. For the automation host, use a supported option such as Home Assistant Green, Raspberry Pi 4 or 5 with at least 2 GB RAM, mini PC, NUC, or virtual machine. Do not treat the Pi memory figure as an LLM requirement.
- Install the model server. Run Ollama on macOS, Linux, or Windows, download the selected model, and verify that it loads without exhausting VRAM or system memory.
- Make the server reachable. Confirm the Home Assistant host can resolve and connect to the Ollama or OpenAI-compatible endpoint. Check firewall rules, listening address, fixed addressing, and sleep settings.
- Configure the integration. Use Home Assistant’s official Ollama integration for Ollama, or the llama.cpp integration when your chosen backend provides the required OpenAI-compatible chat-completions API.
- Expose a small, safe entity set. Begin below 25 entities, preferably with low-risk devices. Enable control only with a model that supports tools, and observe every command during evaluation.
- Measure the real workload. Test a short request, a long entity-aware request, two overlapping requests, and a recovery after the model server restarts. Record latency, memory use, tool-call accuracy, and failure behavior before expanding access.
Optional radios are separate from model compute
If your automations use Zigbee or Thread, a USB radio such as Home Assistant Connect ZBT-2 can provide those protocols; Home Assistant’s FAQ also lists Matter. The adapter does not add VRAM, system RAM, or inference speed. Buy it for the wireless protocol your devices require, not as part of the LLM computer specification.
Common mistakes and how to correct them
Assuming a Home Assistant host is an LLM host
A Raspberry Pi 4 with 2 GB RAM may satisfy the stated Home Assistant installation path while being unsuitable for a chosen model and context. Keep the automation requirement and model requirement in separate checklists.
Using download size as the memory budget
A 7.2 GB model download does not prove that 7.2 GB of RAM is sufficient. Add loaded runtime memory, context cache, concurrent requests, and operating-system overhead, then verify with the model actually running.
Buying a GPU without checking software support
Acceleration depends on drivers, compute capability, operating system, and backend support. Confirm the intended NVIDIA, Metal, or Vulkan path before purchase and budget for the memory the model needs, not merely the GPU’s theoretical speed.
Exposing every entity immediately
Large entity lists consume context and make tool selection harder. Smaller models may issue incorrect commands or lose track of a conversation. Start with a deliberately limited, low-risk set and expand based on observed accuracy.
Ignoring server availability
A model on a desktop that sleeps is not an always-available voice assistant. Disable sleep where appropriate, or move inference to a machine designed to remain online. Test the integration after a reboot and after the model process is restarted.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Expecting an API-compatible server to be behaviorally identical
OpenAI-compatible endpoints standardize request shape, not model quality or Home Assistant tool behavior. Validate tool calls, error handling, authentication, and latency with your exact backend and model.
A practical buying checklist
- Model name, quantization, and documented memory guidance are known.
- Available VRAM or Apple unified memory covers the model plus the intended context, with margin.
- System RAM covers fallback, operating-system overhead, and other services if VRAM is insufficient.
- CPU-only operation has been tested if you do not plan to use a supported GPU path.
- Expected simultaneous requests are included in the memory estimate.
- The LLM server can remain powered on and is reachable from Home Assistant.
- The selected model supports tools if it will control entities.
- The initial entity scope is deliberately small and below Home Assistant’s suggested 25-entity experiment limit.
- Home Assistant’s own host meets its installation requirements independently of the LLM specification.
- Any Zigbee, Thread, or Matter radio is selected separately from compute hardware.
What to expect from the finished system
A well-sized setup is a set of compromises rather than one magic specification. More model memory and a supported GPU can improve responsiveness; longer context and more simultaneous users consume that memory; a separate Home Assistant host improves isolation but adds hardware and power use. Smaller models reduce requirements but are more likely to make mistakes, especially when controlling devices. The safest path is to size for the model and workload you actually intend to run, connect it over a reliable network, and expand control only after measured testing.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
