> **Warning:** This Epic's description is managed as documentation and mirrors a file in the > website repository. Do not edit it directly. See [how we manage the roadmap](https://selorahomes.com/handbook/departments/product/roadmap-management/) > and propose all content changes via a Merge Request. [📝 Source](https://gitlab.com/selorahomes/products/selorahomes.com/-/blob/main/content/docs/roadmap/home-assistant-intent-management/index.md) ## Overview Home Assistant exposes user commands as [intents](https://developers.home-assistant.io/docs/intent_index/) — typed intentions such as `TurnOnIntent`, `SetFanSpeedIntent`, or `QueryWeatherIntent` carrying typed slot values. Every voice assistant (Conversation, Alexa, Google, and the mobile app) funnels through the same intent layer, which makes intents the single integration point between natural language and a Home Assistant installation. Today, intent coverage and quality are fully local to the installation: which intents exist, how slots are resolved, and what happens when a command is ambiguous or unsupported are all determined by the HA build and the entities present. Selora AI takes over that layer: it registers, extends, and curates intent handling per property, giving customers consistent voice and text behavior regardless of the underlying Home Assistant version. ## Customer value - **Consistent commands**: The same phrases work the same way across every Selora-managed home, with predictable fallbacks when something is unavailable. - **Extended coverage**: Selora AI handles phrasings the stock HA intent layer does not support (multi-device commands, conditional phrasing, custom routines), mapping them to the right services and entities. - **Curated slot resolution**: Ambiguous commands ("turn the lights on" in a house with six areas) are resolved using rooms, usage history, and customer preferences instead of failing or guessing. - **Installer oversight**: Installers see which intents fire, how often, and where they fail, so they can tune behavior without touching HA config files. ## Scope - **Intent registry**: Selora AI maintains a per-property registry of supported intents, sourced from Home Assistant's intent index and extended with Selora-specific intents. - **Slot resolution engine**: Maps slot values (area, device, mode, level) to concrete entities using the installation manifest, rooms, and usage data. - **Fallback and clarification**: Detects ambiguous or unsupported commands and responds with clarifications instead of errors. - **Intent analytics**: Per-intent firing, success, and fallback metrics surfaced to installers. - **Custom intent authoring**: Let customers define named routines in natural language and expose them as first-class intents. ## Architecture 1. **Intent listener**: Selora AI subscribes to Home Assistant intent events and intercepts them before default handling. 2. **Registry lookup**: The per-property registry determines whether Selora handles the intent or defers to the default HA handler. 3. **Slot resolver**: Enriches slots against the installation manifest and entity registry to produce a concrete action plan. 4. **Execution**: Actions run through the standard Home Assistant service layer, so all normal permission and state rules still apply. 5. **Feedback loop**: Firing, success, and fallback outcomes are recorded and reported to installers. ## Open questions - **Precedence rules**: When does Selora defer to the default HA intent handler versus handle the intent itself? - **Language coverage**: Which languages get full Selora intent coverage at launch? - **Custom intents**: Should custom routines be exposed to voice assistants immediately, or staged behind a confirmation step?