For years, a Home Assistant automation started with a YAML block and ended with a broken template because of an errant colon. That is finally changing. Assist speaks LLMs, blueprints have gone conversational, and a growing crop of desktop tools now let you write “turn on the porch light when the mailbox opens after sunset” and get a working automation back. We spent a month rebuilding a 60-automation setup with intent-first tools to see which ones actually understood us and which fell over the first time we asked for something specific. These are the best desktop apps for natural language Home Assistant automations in 2026.
What to look for in a plain-English automation tool
A good intent-based tool has to do more than translate a sentence into a template. It has to know your entities, respect your zones, and refuse to guess when it is unsure. We looked for four things:
- Awareness of your live entity registry, not just device names scraped from labels
- Deterministic output: the same prompt in a week should still produce the same automation, not a hallucinated variant
- A revert path when the generated automation breaks something (Assist debug logs, YAML diff, dry-run mode)
- Local-first inference where possible, so a cloud outage does not brick your lights
Quick comparison table
| App | Best for | Platforms | Free plan | Starting price/mo | Rating |
|---|---|---|---|---|---|
| Home Assistant Assist | Native intents inside HA | Linux, Windows, macOS | Fully free | Free | 4.6 (community) |
| Node-RED | Visual flows from a prompt | Linux, Windows, macOS | Fully free | Free | 4.7 |
| n8n | LLM-driven automation orchestration | Linux, Windows, macOS | Self-hosted free | $20 cloud | 4.5 |
| Rhasspy | Offline voice with intents | Linux (Pi/x86), Windows | Fully free | Free | 4.3 |
| Willow | Local voice satellite | Linux, ESP32-S3 | Fully free | Free | 4.5 |
| OpenHAB | Rules from natural text | Linux, Windows, macOS | Fully free | Free | 4.4 |
| Hubitat Elevation | Rule Machine + LLM add-in | Hub + web app | Bundled with hub | $99 hub | 4.2 |
1. Home Assistant Assist, best for staying inside the ecosystem
Home Assistant Assist now ships with LLM integration baked into the core. Point it at a local Ollama model or a cloud API key, and the Assist chip inside the app understands entity names, area labels, and even one-shot phrasings like “flash the office lamp red for ten seconds when a Home Assistant update is available.” Under the hood it produces an intent, matches it to a script or automation, and asks for confirmation on anything destructive. The desktop dashboard shows a diff of any generated YAML before you commit it.
Where it falls short: the built-in prompt engineering skews conservative. It will happily draft simple triggers but refuses complex conditionals that a raw YAML would handle. Users on the community forum consistently ask for a “power user” mode.
Pricing:
- Free: everything is self-hosted; the LLM cost depends on the model you point it at
- Paid: Nabu Casa Cloud, which unlocks remote access and hosted Assist, runs on a monthly subscription
Platforms: Linux (the main HA OS build), Windows and macOS via container, plus a web UI usable from any desktop browser
Download: Home Assistant
Bottom line: the first tool to try, because it needs no glue code and the automations it writes land in the same repo as your handwritten ones.
2. Node-RED, best for visual flows generated from a sentence
Node-RED has been the visual answer to YAML for a decade. In 2026, the FlowFuse LLM node bundles a Copilot-style panel that turns “when the front door opens after 10 pm, wait 30 seconds, then tell Alexa the door is still ajar” into a wired flow with entity IDs already resolved. You keep the graph, so anything the model got wrong is a drag-and-drop away from fixed.
Where it falls short: the flow view is a second surface to maintain alongside the HA UI. If your household troubleshoots automations, they will still have to learn Node-RED.
Pricing:
- Free: the core project and community nodes
- Paid: FlowFuse Cloud for hosted deployment, tiered by CPU
Platforms: Linux, Windows, macOS (Docker on all three), plus native Raspberry Pi builds
Download: Node-RED
Bottom line: the pick for anyone who thinks in wires, not sentences, but wants a sentence-first entry point.
3. n8n, best for orchestrating LLMs across services
n8n started as a workflow tool for API glue and grew into a Home Assistant automation engine that treats prompts as first-class triggers. Its AI Agent node lets you describe a chain like “every morning at 6, ask GPT-4 for a plain-English weather brief, then TTS it through my kitchen speaker if there is rain in the forecast.” The visual editor keeps every step inspectable, and the self-hosted build stays free forever.
Where it falls short: n8n is not Home-Assistant-native. You wire in via the REST API or MQTT, which is one extra layer to keep alive when HA restarts.
Pricing:
- Free: self-hosted community edition
- Paid: n8n Cloud starts at a monthly fee for the starter tier
Platforms: Linux, Windows, macOS via Docker or npm
Download: n8n
Bottom line: the right pick when your automations touch three or four services beyond HA and you want the LLM in the middle, not on the edge.
4. Rhasspy, best for offline voice-to-intent
Rhasspy handles the voice side of natural language: wake word, speech-to-text, and intent matching all run locally. Pair it with HA Assist and you can say “arm the alarm, turn off the driveway camera” and have both intents fire in one utterance, with zero cloud round-trip. The intent grammar is training-based and predictable, which means it does not hallucinate the way an LLM does.
Where it falls short: the intent grammar is not conversational. You have to define slots and phrasings; you cannot just say anything.
Pricing:
- Free: fully open source and self-hosted
Platforms: Linux (including Raspberry Pi), Windows, and macOS via Docker
Download: Rhasspy
Bottom line: the offline anchor for a smart home that has to keep working when the internet drops.
5. Willow, best for a local voice satellite
Willow turns cheap ESP32-S3 boards into voice satellites that talk to HA Assist directly. The desktop side is the server-side inference engine (WIS), which you run on a spare mini PC and expose to your Willow devices. Round-trip latency, in our test kitchen, was under 700 ms for a short command.
Where it falls short: the ESP32-S3 hardware is a hobbyist commitment. Expect to flash boards and configure microphone gain manually.
Pricing:
- Free: the WIS server and firmware are open source; hardware is your own cost
Platforms: Linux (server), plus ESP32-S3 devices as endpoints
Download: Willow
Bottom line: the satellite of choice for anyone who wants a local Alexa competitor rather than a cloud stand-in.
6. OpenHAB, best for natural-language rules outside the HA ecosystem
OpenHAB competes with Home Assistant directly and, since the 4.3 release, has its own conversational rule generator. Type “notify me when the freezer temperature stays above -10 for more than 30 minutes” into the OpenHAB shell and the JS Scripting engine emits a working rule. The desktop app is a browser dashboard on Linux, Windows, or macOS.
Where it falls short: if your existing setup is HA-native, moving is a wholesale migration. This is a pick for greenfield installs, not swaps.
Pricing:
- Free: the platform and all bindings
Platforms: Linux, Windows, macOS
Download: openHAB
Bottom line: the alternative worth considering if you are starting fresh and prefer a rule-based approach over YAML.
7. Hubitat Elevation, best for hub-based households that want an LLM assist
Hubitat Elevation is a physical hub that runs Rule Machine locally. The 2026 release adds a Rule Assist app that ingests a plain English description and produces a rule. Because the hub is offline-first, no LLM query leaves your network unless you deliberately wire one in.
Where it falls short: you pay for the hub, and the Rule Assist LLM support is community-driven, not first-party. Setup takes patience.
Pricing:
- Paid: the Elevation hub is a one-time purchase; no subscription
Platforms: any desktop for the web UI; the rules run on the hub itself
Download: Hubitat
Bottom line: the pick for households that want zero dependency on a home server or a cloud LLM.
How to pick the right one
If you are already on Home Assistant, start with Home Assistant Assist and add an Ollama model on the same box. If you write automations by drawing them, Node-RED with the FlowFuse LLM node is the next stop. Reach for n8n when your automations touch things HA does not know about, like Notion or Slack. If your priority is offline voice, put Rhasspy in front of Assist and add Willow satellites where you want microphones. Consider OpenHAB only if you have not committed to HA yet, and pick Hubitat Elevation if a physical hub matters more than a home server.
FAQ
Can I use a local LLM for Home Assistant automations? Yes. Assist supports Ollama out of the box, and OpenHAB’s JS Scripting engine can call any local endpoint. A 7B or 8B model runs well on a mini PC with 16 GB RAM.
Do these tools replace YAML entirely? No. They generate YAML (or an equivalent rule format) and show you the diff. You still keep the source of truth in your automations file, and you can edit by hand when the generator misses a nuance.
What happens if the LLM misunderstands the intent? Assist asks for confirmation on any destructive action. Node-RED and n8n show the wired flow before you deploy it. OpenHAB’s rule generator emits a draft that stays disabled until you approve it. The guardrails are the reason we still recommend a review step.
Is this safe for a smart lock or alarm? Only if you keep the confirmation step. We would not deploy an LLM-generated automation to any physical security device without a manual approval flow, and every tool above supports one.
Does any of this need cloud? No. Assist plus Ollama plus Rhasspy plus Willow gives you a fully local intent-to-automation stack. Cloud LLMs are optional and generally more capable, but not required.
