Selora AI: Create Scenes

Overview

Enable users to create Home Assistant scenes by describing what they want in natural language. Selora AI discovers the relevant devices from the current conversation context — across domains (lights, media players, covers, etc.) — and assembles the scene automatically.

When context is not sufficient, Selora AI suggests matching devices or asks clarifying questions to build the scene collaboratively.

Example

User: “Create a movie night scene”

Selora AI discovers light.living_room_ceiling, light.living_room_lamp, media_player.tv, and cover.living_room_blinds from the conversation context because they are part of the current discussion. It then proposes a scene that dims the lights, turns on the TV, and closes the blinds.

If the context is ambiguous, Selora AI asks: “Should I include the kitchen lights as well?” or suggests devices: “I found these devices in your living room — which ones should be part of the scene?”

Customer value

• Natural interaction: Users describe scenes in plain language instead of manually picking entities and setting states.
• Context-aware discovery: Selora AI leverages the ongoing conversation to identify relevant devices across multiple domains without requiring the user to list them.
• Guided creation: When context is not sufficient, Selora AI suggests devices or asks targeted questions — users are never left guessing.
• Cross-domain support: Works across all supported Home Assistant domains (lighting, climate, media, covers, etc.).

Scope

• Conversational scene builder: Intent extraction for scene creation and state mapping.
• Context-based device discovery: Logic to identify relevant devices from recent conversation history.
• Interactive fallbacks: Collaborative UI/UX for disambiguation and device selection.
• Preview and confirmation: Show the proposed scene states before saving.
• Multi-domain support: Integration with core Home Assistant entity registries.

Architecture

1. Intent extraction: Selora AI identifies the “create scene” intent and extracts the desired mood or goal.
2. Device resolver: Scans conversation context for recently mentioned or interacted-with devices.
3. Entity filter: Matches extracted devices against the user’s Home Assistant registry.
4. State mapper: Maps high-level intents (e.g., “dim”, “close”, “on”) to concrete entity states.
5. Confirmation: Selora AI presents the plan to the user.
6. Persistence: Saves the scene via the Home Assistant scene API.

Open questions

• State inference: How to reliably infer specific states (e.g., “movie night” brightness) without explicit user input?
• Disambiguation: Best UI pattern for selecting between multiple similar devices in a chat interface.
• Context boundaries: How far back in the conversation should the device resolver look?