Hi everyone, I’m building a multi-step automation in Home Assistant and I’d like some advice on the best way to structure it.
Here is the workflow I want to achieve:
- 7:00 – Turn off the water pump
- 7:05 – Turn on the fish feeder for 1 minute, then turn it off
- 7:06 – Trigger a camera to take a snapshot of the feeding area
- Send the snapshot to an AI Vision service (OpenAI Vision, Gemini, Frigate+LLM, etc.) to check:
- Whether food has actually dropped into the water
- Optional: whether fish are gathering around the feeding spot
- Based on the AI result, send me a notification (e.g., “Feeding successful” / “No food detected”)
- 7:10 – Turn the water pump back on
My questions:
- What is the best practice to implement this workflow in Home Assistant?
- Should I break it into multiple automations, scripts, or one large automation?
- What is the most reliable way to trigger camera snapshots at a precise time (Frigate vs. direct camera integration)?
- For image analysis, should I use AI Task, LLM Vision, Frigate, or call an external API?
- How do you recommend handling the AI result (JSON structure, variables, conditions) inside HA automation?
Any tips, example YAML, or recommended architecture would be greatly appreciated.
I want this routine to be stable and accurate because it’s part of a daily smart-aquarium setup.
Thanks in advance!