AI
Every model used, what it does, and why. No invented capabilities — this is exactly what runs behind each guest interaction.
Chat & reasoning
Each guest message triggers a retrieval step that finds the most relevant sections of your knowledge base, then passes them to GPT-5-nano to generate a response. The model is instructed to stay within what it knows about your property — it won't guess.
Knowledge base retrieval
Every knowledge base section is embedded as a dense vector when you create or update it. At inference time, the guest's question is embedded and compared against all stored vectors using cosine similarity. The top-matching sections are injected as context.
Real-time voice
The voice assistant uses OpenAI's Realtime API over WebRTC. When a guest taps the microphone, the client connects directly to a session with a server-issued ephemeral token. Audio and text modalities are both active — meaning the transcript is also stored.
Content translation
Property owners can translate their entire knowledge base, amenities, house rules, and room descriptions to any supported language with one click. The translation runs as a batched JSON operation for efficiency.
The concierge is instructed to answer only from your property's knowledge base. It does not have internet access, does not know other properties, and does not hallucinate answers it cannot support. When uncertain, it tells the guest to contact the host directly.