Hi,
I’ve been wanting Grok inside Homey for a while, so I built a small app for it. It’s called Grok Connect. Unofficial, not made by xAI, and Homey Pro only.
You put a card in a Flow, ask something, and get an answer you can actually branch on. Yes or no, a number, a choice from a list you wrote, or just a sentence.
It can search the web or X, look at a camera snapshot, and keep a short conversation. You use your own key from console.x.ai, under API Keys. A normal Grok subscription doesn’t work here. There’s a daily cap, so a Flow can’t quietly burn through your credit.
It’s in the Homey App Store now: Grok Connect | Homey
You install it like any other app, then paste your own xAI API key from console.x.ai. A Grok subscription is not a key.
If you try it and a card feels awkward, just say so.
Djordie
A few Flows I use it for. Your devices will be named differently, obviously.
Washer. At 9 the Flow asks if now is a decent moment. I pass the current price, the cheap hour, how much solar is coming in, and whether rain is expected. Confidence at least 70%. If Grok isn’t sure, it does nothing.
Doorbell. Someone rings, the snapshot goes to Grok with “who or what is at the door, mention parcels”, and the living room speaker says the answer.
Morning. Alarm goes off. Grok writes at most four sentences: weather, anything notable, and whether I need an umbrella. Web and X search on. Then I run that through “make speakable” and the kitchen speaker reads it.
Small update: 1.1.0 is live in the store now.
New cards: ask for fields, a list, a date or time, a score, compare two options, split an answer, and suggest a device or scene. That last one only hands you a name back. It doesn’t switch anything itself, your Flow decides what happens.
One thing that trips people up: you need an API key from console.x.ai (API Keys, it starts with xai-). A Grok subscription is not the same thing and gets rejected. The xAI account also needs a bit of credit.
Store: Grok Connect | Homey
If you try it, I’d like to hear which card you used and what felt clunky.
Nice idea, and since everyone brings their own key, a rough per-call figure may help people pick the daily cap: xAI’s current card has Grok 4.7 at $2 per million input tokens and $6 per million output (reasoning tokens bill as output), and Grok 4.3 at $1.25 / $2.50. A washer-style yes/no call with about 500 tokens of context (price, cheap hour, solar, rain) and about 300 tokens out including reasoning comes to roughly 0.28 cents on 4.7 and 0.14 cents on 4.3, so ten of those a day stay under 3 cents. The cards that cost more are the ones that make the prompt big, like a long conversation or a lot of search results pulled in, and on 4.7 a prompt of 200K tokens or more doubles every rate for the whole request. If the app sends the same instruction prefix every time, cached input on 4.7 is $0.50/M instead of $2 (xAI suggests setting a prompt cache key so the hits are reliable). Is the model selectable per card? For simple yes/no gating, 4.3 at less than half the output price looks like a sensible default.
@Djordie_Hendriks Thanks, that’s exactly what I had in mind. Being able to set the model per card is the best part for me: a cheap model for the yes/no checks and the big one only where the answer actually needs reasoning.
I don’t run Homey myself, so I can’t tell you whether existing Flows survive the update. One thing I noticed: the store page already shows 1.3.0, not 1.2.0.
Two cost ideas. Could the usage stats show cached input tokens next to the normal ones? Then people can see whether the cache key is actually kicking in, and it works best when the changing tokens (price, weather, time) go at the very end of the prompt. And if Live search and Imagine calls are billed apart from tokens, it would be good to count them in “Spending today” too, otherwise the daily limit will run a bit low.
Hi,
Thanks, good to hear the per-card model is the part you like.
You’re right about the version. I skipped ahead to 1.3.0, which also added a few new cards (image checks, notes in a conversation, a digest summary, image generation and a spending trigger).
Your cost ideas are in 1.3.1, which is going to test now:
- The usage stats show input tokens with the cached part and a percentage, plus output tokens with reasoning, both for today and in total. That way you can see if the cache key is kicking in.
- In chats with memory, the time sat before the history and broke the cache every minute. It now goes at the very end of the last message, so the instructions and history stay the same between requests.
- “Spending today” already used the exact cost xAI returns with every response, which includes Live Search and Imagine, so the daily limit was accurate. The backup estimate for when that figure is missing now also follows xAI’s current search pricing.
Thanks again for thinking along, it really made the app better. 
Djordie
