ZimaBoard 2: Frigate, Ollama and Home Assistant setup

This setup and review tests ZimaBoard 2 as a local AI smart home server with Ollama, Home Assistant and Frigate. The speaker reports mixed results: Frigate runs well, but language model responses can take too long for voice control or urgent camera alerts.

The walkthrough installs Ollama through the board's app store, downloads Qwen 2.5 and connects Home Assistant using an IP address and port without an API key. Enabling Assist sends the Home Assistant context to the model, which the speaker finds slow on this hardware. Letting Home Assistant handle local commands while the model answers general questions works better in the demonstration, though the speaker still reports about a 15-second delay.

Frigate installation uses a custom Docker Compose YAML file. The speaker then configures MoonDream to describe camera snapshots and builds a Porch Patrol automation. It fetches a detected person's image, resizes it, requests a description through an AI task, sends a phone notification and records the event in a local log.

Resizing reduces the reported notification wait to roughly 40 to 45 seconds in some tests, but another trace shows 57 seconds for image processing. The speaker recommends this hardware for experiments and tasks that can wait, rather than primary doorbell alerts.