Hopping see integration of Homey and OpenClaw or other local AI in the future. Anyone know if someone already do something with this?
You can use Ollama in your Homey flows, you can also ask it to analyze camera snapshots for example.
bit struggeling with ollama, it has âthenâ cards, but how to use the model results in a flow?
i want to obtain weather advise (preferrably using agent) to control blinds,sunscreens and lighting
openclaw integrates with ollama (or directly to the cloud based llmâs), but i think you still need the right agents set.
@Vnet I ran into the same thing: the Ollama cards give you text, and then youâre stuck turning that into a decision in a Flow.
I made an app for this, Cortex AI (so Iâm biased). It works with your local Ollama, and its cards return something a Flow can use directly: a yes/no And card, one of your own options, or a number. For your blinds youâd put the weather, the sun times and the blinds in a âbrainâ, then use âLet Cortex choose one optionâ with options like open, half and closed, and do the rest with normal Flow cards. If the model is slow, the card falls back to a value you choose, so the Flow never hangs.
More in the app topic: [APP][Pro] Cortex AI â an AI brain for your Flows and your home
@calvarezh OpenClaw speaks MCP, so you can connect it to Homey directly. I made an app for that, MCP AI Bridge (so Iâm biased too): it runs an MCP server on the Homey itself, and OpenClaw, or any other MCP client, can then read and operate the whole house: devices, flows, moods, energy, and so on.
In OpenClaw it is one server entry:
openclaw mcp add homey --url http://<homey-ip>:52199/mcp --transport streamable-http
openclaw mcp doctor homey --probe
The app makes an API key on its Status tab. Give OpenClaw its own key on the Security tab, with Control only if it should switch things but not change flows, and add it as an Authorization: Bearer <key> header in OpenClawâs MCP settings rather than in the JSON. With a local model, switch that key to compact mode or pick a few tools: the bridge has close to 200, and small models get lost in that many.
@Vnet for the blinds: the appâs âAsk a modelâ cards work with Ollama and LM Studio too. In the version in the App Store now you get the answer as text, plus a yes/no condition. 5.2.1, which is with Athom for review, adds cards that return one of your own options (open, half, closed), a number between two values, or a yes with a minimum confidence, each with a time limit and a fallback so the flow never hangs. The weather and sun times can go along as context.
More in the app topic: [APP][Pro] MCP AI Bridge: talk to your Homey with Claude, ChatGPT, Gemini, Copilot, Cursor, Mistral, Ollama, LM Studio, n8n, Home Assistant or any AI, also from Telegram and WhatsApp