Choosing Your Hardware.

Why the default recommendations don't fit serious agriculture.

Most introductory Home Assistant resources recommend a Raspberry Pi. This is reasonable advice for a specific audience — someone exploring smart-home automation as a hobby, willing to buy a new device, content with the capacity the Pi provides, and unlikely to push the system into territory that taxes it. That audience exists and is well-served by a Pi.

Serious agricultural deployment is a different situation. A greenhouse monitoring system is operational infrastructure, not a hobby. It runs continuously, through power events and environmental stress, across seasons. The grower depends on it. The natural growth arc of such a system includes more sensors over time, longer historical data, eventually an analytics layer (Grafana and InfluxDB), eventually computer vision (Frigate on camera feeds), eventually local AI (Ollama). A Raspberry Pi handles the starting point competently and chokes progressively as the operation grows. By the time the grower wants to add cameras with object detection, the Pi is either limiting or being replaced.

The alternative — starting with a more capable machine — costs almost nothing if the grower uses repurposed hardware, and costs a moderate amount if they buy used or new business hardware. The same dollars produce far more capacity than purpose-built Home Assistant hardware offers at comparable prices. And because the machine is general-purpose, it also runs Docker containers for every other useful service the grower might want alongside Home Assistant — the graybox approach that this section returns to repeatedly.

This is not a criticism of the Pi or of Home Assistant's purpose-built hardware. Both serve their audiences well. But the audience this site addresses — growers running working agricultural operations — is generally better served by hardware chosen for capability and longevity rather than compactness and simplicity.

The tiered recommendation.

The collective's hardware recommendation comes in three tiers, roughly in the order of what most agricultural deployments should reach for first.

### Primary: Repurposed business-class desktops and laptops.

An old office desktop — an ex-corporate Lenovo ThinkCentre, HP EliteDesk, HP ProDesk, Dell OptiPlex, or Dell Latitude — has been engineered for years of continuous service in business environments. The processors are Intel Core i5 or i7 from recent enough generations to handle modern workloads. The motherboards support substantial RAM (often 16 to 64 GB) and multiple storage devices. The chassis designs accommodate cooling, cable management, and expansion. And when a business cycles these machines out of their fleet — typically every three to five years — they often become available cheap or free.

A business-class laptop — a ThinkPad T-series, T440 onward; an HP EliteBook; a Dell Latitude 7000-series — has all of the above plus a built-in battery that serves as a free uninterruptible power supply. When utility power blinks (and it blinks a lot in rural locations, during storms, during heat waves when the grid is stressed), the laptop keeps running on its own battery for an hour or longer. No external UPS required for that function. A grower with a working ThinkPad in a drawer has, essentially, a Home Assistant platform already.

Typical capability from repurposed hardware: a ThinkCentre from around 2018, acquired for free or cheaply, might have an Intel Core i5-8500T, 16 GB of RAM, a 256 GB SSD, and plenty of USB ports. That is vastly more than a Raspberry Pi 5 offers. The same machine handles Home Assistant, Frigate with several cameras, InfluxDB with years of history, Grafana for dashboards, Mosquitto as the MQTT broker, and several ESPHome compile jobs per week, all simultaneously, with room to spare. It will continue serving for another five to ten years before reaching end of life.

When this tier fits: Almost every operation that takes monitoring seriously. Small farms, commercial greenhouses, packhouses, multi-site operations, research stations. The capacity headroom and the graybox flexibility fit a wide range of situations.

When this tier does not fit: Operations with strong preferences for a small form factor, operations unwilling to administer a general-purpose Linux system, operations where the machine must run in an environment unsuitable for a desktop chassis (extreme heat, heavy dust, vibration). For those situations, one of the other tiers may fit better.

### Secondary: Refurbished or new business-class mini PCs.

When repurposed hardware is not available — the grower has no access to old office machines, no family members upgrading, no local business resellers — a refurbished or new mini PC from a business brand is the next best option. These are the desktop equivalent of the repurposed route but in a small form factor.

Specific lines worth looking for: Lenovo ThinkCentre M-series in the Tiny form factor (M70q, M75q, M710q, M720q, M910q, M920q variants depending on the year). HP EliteDesk Mini (800 G3, G4, G5 Mini lines; 600 G-series). Dell OptiPlex Micro (3080, 5080, 7080 Micro lines). Intel NUC in the business-grade models. These machines combine compact size — typically about the size of a hardcover book — with business-class reliability and capability.

Refurbished typical pricing: a moderate amount depending on specifications and condition. New typical pricing: a moderate amount for comparable configurations. Either way, the capability-per-dollar substantially exceeds purpose-built Home Assistant hardware. A moderately priced refurbished ThinkCentre M720q Tiny offers more capacity than an affordable Home Assistant Yellow.

When this tier fits: Operations where repurposed hardware is not available, operations with strong preferences for small form factor, operations where new-with-warranty matters (for institutional settings, for operations that cannot tolerate downtime while diagnosing unknown refurbished machines). Also a good choice as a spare — a new mini PC kept ready as a hot or cold standby for the primary Home Assistant machine.

When this tier does not fit: Operations with access to good repurposed hardware (the savings and the avoided waste argue for the repurposed path). Operations that genuinely need the simplicity of an appliance (those readers may be better served by the tertiary tier).

### Tertiary: Purpose-built Home Assistant hardware.

Home Assistant Yellow, Home Assistant Green, and the Raspberry Pi ecosystem form the tier of purpose-built hardware. These machines are designed specifically for running Home Assistant (in the case of Yellow and Green) or for general hobbyist computing (the Pi), with Home Assistant OS support as a standard path.

Home Assistant Green is a dedicated smart-home hub (built on a Rockchip RK3566 SoC) sold by the Home Assistant project in a consumer-friendly package, typically affordable with Home Assistant pre-installed.

Home Assistant Yellow is a more capable device, also sold by the project, with a CM4 compute module, a Zigbee/Thread radio built in, PoE+ support, and an NVMe slot. Typical pricing a moderate amount depending on configuration.

A Raspberry Pi 4 (typically 4 GB or 8 GB of RAM) or Raspberry Pi 5 with a case, power supply, and storage device costs a modest amount depending on the configuration. The Pi is ubiquitous, well-documented, and supported by a vast community.

When this tier fits: Readers exploring Home Assistant as a hobby, operations that want the absolute simplest path from purchase to running system, deployments where the monitoring scope will remain small, readers for whom the appliance-like nature of Home Assistant OS on purpose-built hardware is valuable, backup devices kept ready for emergencies.

When this tier does not fit: Serious agricultural deployments with an expected growth path into AI, computer vision, substantial history, or multipurpose services. These capabilities eventually tax the tier-3 hardware, and the grower who starts there frequently ends up replacing it. Starting with a more capable machine avoids the rework.

The graybox approach.

A theme that runs through these tiers is the graybox approach — a single capable machine running Home Assistant alongside other useful services, all in Docker containers, all under the grower's control.

A typical graybox deployment runs Home Assistant, Mosquitto (MQTT broker), MariaDB (external database for Home Assistant's history), ESPHome (for compiling and flashing custom devices), Zigbee2MQTT (if the operation uses Zigbee), Frigate (if the operation uses cameras with object detection), InfluxDB and Grafana (for analytics), possibly Node-RED (for flow-based automation), possibly Ollama (for local AI), a backup tool, and possibly a VPN endpoint (WireGuard or Tailscale). All running simultaneously on one physical machine, separated into Docker containers for isolation and ease of management.

The graybox approach works on any machine with enough capacity — tier-1 and tier-2 hardware both support it comfortably, tier-3 hardware does not. This is the central practical reason the collective recommends the capable path. A grower who installs Home Assistant OS on a Raspberry Pi has made a decision to run Home Assistant only on that machine. A grower who installs Home Assistant Supervised on Ubuntu running on a repurposed desktop has made a decision to run Home Assistant alongside everything else they will want over the next five or ten years.

The [Installing Home Assistant on Ubuntu with Docker](/home-assistant/installation/ha-docker) page covers the specifics of setting up the graybox. The [Repurposing a Desktop Computer](/home-assistant/hardware/repurposed-desktop) page walks through acquiring and preparing the hardware.

Sizing considerations.

The specific capacity a grower needs depends on the scope of the deployment. Rough guidelines by operation size.

### RAM.

4 GB supports a basic Home Assistant deployment with a modest number of entities and simple automations. This is the lower bound for a usable system. Fits a Raspberry Pi 4 with 4 GB configuration, a Home Assistant Green, or an old tiny PC with limited expansion.

8 GB comfortably handles Home Assistant plus a few add-ons (Mosquitto, ESPHome, a small database). Fits most small operations well. Available on Raspberry Pi 4/5 with 8 GB, Home Assistant Yellow, entry-level mini PCs, and most repurposed business machines.

16 GB handles the graybox pattern — Home Assistant plus Mosquitto plus a database plus ESPHome plus Zigbee2MQTT plus Grafana plus InfluxDB — with comfortable headroom. Standard on many business desktops from recent generations. A sweet spot for most agricultural deployments.

32 GB or more supports the full graybox including Frigate with multiple cameras, Ollama with meaningful local LLM models, and headroom for growth. Available on higher-tier business machines and most workstation-class hardware. Worth it for operations with ambitious plans.

A consideration: Home Assistant itself is relatively modest in RAM use (usually under 2 GB), but the add-ons around it collectively consume much more. A grower who runs Frigate with four cameras at high resolution easily uses 4 to 8 GB for Frigate alone. A grower who runs Ollama with a meaningful LLM (7B or 13B parameter model) uses substantial memory. RAM is the most common limiting factor on older hardware; upgrading it is usually cheap if the motherboard supports more than is currently installed.

### Storage.

64 GB is the absolute minimum, appropriate only for purpose-built hardware running Home Assistant OS with no substantial history, no cameras, and no analytics.

256 GB SSD handles Home Assistant with reasonable history (months to a year), modest add-ons, and some Frigate recording at modest retention. Minimum for a serious deployment.

500 GB to 1 TB SSD handles the graybox with Frigate recording, Grafana/InfluxDB analytics, years of history, and reasonable headroom. Sweet spot for most operations.

2 TB or more supports long retention, substantial Frigate storage, large LLM models (Ollama stores can run tens of gigabytes), and multi-site configuration. Worth it for ambitious deployments.

SSDs are strongly preferred over spinning disks for the operating system and Home Assistant's database. Spinning disks still have their place for bulk storage (Frigate recordings, long-term backups) where capacity per dollar matters more than access speed. A common pattern: a smaller SSD for the OS and active data, plus a larger HDD for Frigate recordings. Most business desktops accommodate both.

### CPU.

Home Assistant's base CPU demand is modest — an Intel Core i5 from several generations ago handles the core platform and most add-ons without stress. The CPU-hungry components are Frigate (particularly with AI object detection), computer vision in general, large LLM inference on Ollama, and heavy analytics processing. For these, more CPU helps substantially.

A practical rule: an Intel Core i5 of any generation from 2018 onward handles everything most growers need. An Intel Core i7 or Ryzen 7 provides headroom for ambitious AI and vision workloads. Older CPUs (first-generation or second-generation Core i-series) work for basic Home Assistant but struggle with modern AI add-ons. An Intel Atom or low-end ARM processor (like the Raspberry Pi's) handles basic deployments but limits what the system can grow into.

Hardware-accelerated video transcoding (Intel Quick Sync, in most Intel CPUs with integrated graphics) matters substantially for Frigate — it lets a modest CPU handle many camera streams that would otherwise require a much more powerful processor. Any Intel Core i3, i5, or i7 with integrated graphics from about 2015 onward has Quick Sync. Ryzen CPUs without integrated graphics (most desktop Ryzens other than the G-series) do not have Quick Sync. This is a consideration if Frigate is in the plan.

A separate AI accelerator — a Coral USB device at a Hailo PCIe card or an NVIDIA GPU at varying prices — dramatically speeds up AI object detection in Frigate and enables meaningful local LLM inference. Not required but valuable for AI-heavy deployments.

### Network.

Wired gigabit ethernet is preferred for the Home Assistant host. WiFi works but introduces unnecessary points of failure for a machine that typically sits in one place. A business desktop or mini PC typically has gigabit ethernet built in. A Raspberry Pi 4/5 has gigabit ethernet. Home Assistant Yellow has gigabit with PoE+ as an option.

For deployments with distributed sensors across a large property, a separate discussion applies — LoRaWAN, mesh WiFi, or ESPHome BLE proxies extend connectivity to remote sensors. The Home Assistant host itself should have a solid wired connection to the local network; that is where the graybox lives.

Power resilience.

A monitoring system that fails during power events is unhelpful precisely when it is most needed — storms, heat waves, grid stress, agricultural equipment starting up. Power resilience matters.

A laptop has a built-in UPS. This is the single best feature of a repurposed laptop for Home Assistant duty. When utility power blinks, the laptop keeps running on battery. A good laptop battery sustains an hour or more of runtime at idle load, which covers most brief outages. Longer outages either need a larger external UPS or an informed shutdown procedure.

A desktop needs an external UPS. A cheap consumer UPS holds a low-power desktop for 15 to 30 minutes. Longer runtime requires a larger UPS or one with an extended battery pack. Many UPS units expose their status through USB, which Home Assistant can read (through the NUT integration or similar), so the system knows when it is running on battery and can behave accordingly — perform a clean shutdown before the battery runs out, alert the grower, stop non-essential processes.

A mini PC or a Raspberry Pi can run off a small UPS or even a USB battery. The low power draw of these devices means a modest battery sustains them substantially. Some Raspberry Pi users run their Pi directly from a USB power bank with a pass-through charging arrangement that provides ride-through during outages.

Solar and battery systems. An operation running off-grid or with substantial solar and battery capacity often has better power resilience than grid-connected operations. Integrating the solar/battery system into Home Assistant (through Modbus or appropriate integrations) lets the system know its own power state.

The [Understanding Power](/fundamentals/understanding-power) Fundamentals lesson covers power infrastructure more broadly. The [Backup and Recovery](/home-assistant/operations/backup) page covers what happens when power does cut out unexpectedly.

Physical environment.

The Home Assistant machine does not belong in the greenhouse. Condensation, high humidity, and temperature swings degrade computer hardware quickly. The machine belongs in a reasonably clean, dry, temperature-controlled space — an office, a utility room, a climate-controlled packhouse area, or a purpose-built equipment room.

A reasonable enclosure matters if the space is not ideal. A small wall-mounted equipment cabinet protects a mini PC from dust and casual damage. Ventilation matters inside the enclosure — an inexpensive fan or passive vents keep the temperature from climbing. For a repurposed desktop in an office environment, no enclosure is needed.

For very rugged agricultural environments — a working barn, a packing shed, an unheated outbuilding — the deployment patterns diverge. A small ruggedized PC in an enclosure, connected by ethernet to the sensing areas, is one approach. A Raspberry Pi in an industrial enclosure is another. Extreme environments may justify purpose-built industrial computers designed for such conditions, which are typically more expensive but engineered for the abuse. The collective's general recommendation is to keep the Home Assistant machine somewhere reasonable and bring sensor data to it rather than trying to put the machine in the harsh environment.

What to avoid.

A few hardware paths that look tempting but generally do not fit well.

Consumer NAS devices as Home Assistant hosts. A Synology or QNAP NAS can run Home Assistant through Docker, and some growers do this successfully. The drawback is that the NAS is optimized for storage rather than for running applications; performance is usually worse than a comparable general-purpose PC, the OS is proprietary (counter to the open-source principle that threads through this site), and support for Home Assistant on NAS hardware is always somewhat second-class. If the operation already has a NAS for storage, using it as the Home Assistant host is a reasonable shortcut; acquiring one specifically for that purpose is not recommended.

Very old hardware. A computer from 2005 with a single-core processor and 2 GB of RAM is not a useful Home Assistant host, even if it is free. The power consumption alone (an old machine running continuously costs more in electricity than a new mini PC over a year or two) makes it a false economy. The rough cutoff: Intel Core i-series (any generation) or equivalent AMD, 4 GB of RAM minimum, SSD capability. Below that, the machine is not worth deploying.

Hardware without ECC memory for truly critical use. Standard consumer and business hardware uses non-ECC memory, which works fine for most purposes. Operations where data integrity is genuinely critical (some research installations, some compliance-sensitive deployments) may justify ECC memory and a workstation-class machine. For most agricultural monitoring, this is overkill. Worth knowing the option exists.

Gaming PCs. A gaming PC has more than enough capacity to run Home Assistant, but is typically optimized for short bursts of high performance rather than continuous reliable operation. Gaming motherboards, gaming RAM, and gaming cooling can handle continuous use but are not designed for it the way business hardware is. A business desktop is a better long-term Home Assistant host than a gaming PC, even if the gaming PC has higher peak performance.

Custom-built PCs from consumer parts. A skilled builder can produce an excellent Home Assistant host from consumer components. For most growers, the effort exceeds the value — a refurbished business mini PC is cheaper, already built, already tested, and comes with a small warranty. Custom builds become worthwhile for specific requirements (a particular GPU for AI work, a particular storage configuration, a specific form factor) that a business machine does not offer.

A decision framework.

The actual decision for a specific grower comes down to a few questions.

Does the grower have access to repurposed business hardware? If yes, start there unless the specifics rule it out. This is the best outcome for cost, environmental impact, and capacity. Evaluate the specific machine against the sizing guidelines above; if it meets the 8 GB RAM, 256 GB SSD, Intel Core i5-or-better bar, it is likely a good choice.

If not, can the grower spend a moderate amount on a refurbished business mini PC? If yes, go there. A well-chosen refurbished machine from a quality business brand provides excellent Home Assistant capacity for modest money. Prefer Lenovo ThinkCentre Tiny, HP EliteDesk Mini, Dell OptiPlex Micro, or Intel NUC. Look for 16 GB RAM (or upgradeable to 16 GB), a 256 GB or larger SSD, and an Intel Core i5 or i7 from the 7th generation or newer.

If a moderate amount for new business hardware fits the budget better, the same families sold new are excellent. More warranty coverage, known history, newer components. The same model selection guidance applies.

If the grower specifically wants the simplest path from purchase to running system, purpose-built Home Assistant hardware is a valid choice. Home Assistant Green is appropriate for very small deployments; Home Assistant Yellow for slightly larger ones. Accept that this path limits what the system can grow into.

If the grower has a Raspberry Pi already and wants to learn Home Assistant, starting there is fine. Migration to more capable hardware later is straightforward — Home Assistant's backup and restore functionality preserves the configuration across hardware changes.

The rest of this sub-section covers each path in depth. [Repurposing a Desktop Computer](/home-assistant/hardware/repurposed-desktop) walks through the primary path. [Repurposing a Laptop](/home-assistant/hardware/repurposed-laptop) covers the laptop variant. [Using a Mini PC](/home-assistant/hardware/mini-pc) covers the secondary path. [Purpose-Built Home Assistant Hardware](/home-assistant/hardware/dedicated) covers the tertiary path. Reading the specific page for the path chosen is the next step.