v3.20.0

Added

• Camera descriptions in your own language, and your own house rules for the camera analyst — two new optional fields in the integration's Options flow, right below the system prompt. Camera description language (e.g. Czech) makes every VLM camera description — the chat camera tool, the save_and_analyze_snapshot service, and the proactive video analyzer — come back in that language instead of hardcoded English. Additional camera analysis instructions is free-form multiline text appended to the VLM system prompt for scene-specific guidance like Ignore cars in the driveway. Both are appended after the built-in prompt rules, never replacing them — though where your extra instructions conflict with or narrow the built-in description request, your instructions win (field testing on chat-tuned VLMs showed the instruction must also ride on the per-image request itself, not just the system prompt, to be honored; live in-chat "check the camera for X" requests always keep priority over stored instructions) — so the repeated-scene detection from #493 stays intact: the internal Scene unchanged. reply is explicitly carved out to stay in English, word for word, because it is matched by code rather than shown to users. Both fields default to empty — behavior is unchanged unless you opt in. Designed, implemented, and field-tested with Czech by @hruba202 in #497.

Full changelog: 3.19.1...3.20.0