I built an open-source, self-hosted take on ElevenLabs' Reception.ai
说明与用途
Frontdesk.ai 是一个开源、自托管的 AI 接待员,替代付费的 Reception.ai。它能接听电话、预约、回答常见问题,并将转录、录音和摘要本地存储,完全可自定义语音、模型和工具。
问题
Dental clinics miss appointments because phones ring while staff are busy and no one answers.
结果
Frontdesk.ai provides a self‑hosted, open‑source AI receptionist that eliminates missed calls, automates appointment booking, answers FAQs, and stores all call data locally, eliminating subscription costs and enabling full customization.
工作流
- 1. User runs
docker compose upto start the stack. - 2. Browser requests a LiveKit room and token from the server.
- 3. Voice worker joins the room and streams audio via WebRTC.
- 4. Voice worker runs Groq Whisper for speech‑to‑text, then a Groq‑hosted LLM with tools
check_availability,book_appointment,search_knowledge. - 5. LLM responses are validated with Zod before database writes.
- 6. On hangup, transcript, recording, summary, and outcome are saved to Supabase.
- 7. User can paste a website URL to auto‑populate business profile and services.
证据
Frontdesk.ai is an open-source AI receptionist that picks up instead. It answers the call, books the appointment against a real calendar, answers questions from the business's own website, and leaves a transcript, a recording and a summary behind.
It runs on your own system. One docker compose up on your laptop or your server. Your calls, clients and recordings live in your own Supabase project.
The agent has three tools: check_availability, book_appointment and search_knowledge. Two rules apply to every tool call: Arguments are validated with Zod before anything is written. The business is resolved on the server, never from an ID the client sends.
No inbound phone calls yet. Calls happen in the browser today. Real phone numbers are the top item on the roadmap.
- 记录日期: 2026-10-10T12:40:55.934Z