ezRental
LIVEezRental 不是「租屋網站」。它處理的是租屋現場最容易拖垮人的部分：重複提問、預約來回、物件資訊散落，以及房東永遠要在不同工具之間切換。ezRental is not just a rental website. It handles the part of renting that drains people: repeated questions, booking coordination, scattered property data, and landlords constantly switching between tools.
這個專案的切法，是把一個真實、瑣碎、跨工具的人工流程拆開，重新組成一套租客與房東都能使用的產品。重點不是「有 AI」，而是 AI 被放在正確的位置：先處理重複溝通，處理不了再交回人。The angle here is taking a real, messy, cross-tool manual process and rebuilding it into a product used by both tenants and landlords. The point is not “it has AI”; the point is that AI is placed in the right layer: handle repeat communication first, then hand off to a human when needed.
為什麼我做這個Why I built this
我做 ezRental，是因為租屋流程裡有太多事情其實不該靠人一直手動接。租客問的問題重複、看房時間來回確認、房源資訊散在不同工具裡，房東最後變成客服、行事曆管理員和資料維護員。這種流程如果不被產品化，只會一直靠人硬撐。I built ezRental because too much of the rental workflow is held together manually. Tenants ask repeated questions, viewing times require back-and-forth confirmation, and property data is scattered across tools. The landlord becomes customer support, calendar manager, and data maintainer. If this workflow is not productized, people keep carrying it by hand.
它解決什麼問題The problem it solves
租屋流程表面上是刊登物件，實際上是資訊同步問題。租客想知道可不可以養寵物、什麼時候能看房、照片是不是最新；房東要維護物件資料、行事曆、對話紀錄、Q&A，還要記得誰問過什麼。ezRental 把這些碎片收斂成一條可管理的流程。A rental workflow looks like listing a property, but it is really an information-sync problem. Tenants ask about pets, viewing times, and whether photos are up to date. Landlords maintain property data, calendar slots, conversations, and Q&A while remembering who asked what. ezRental turns those fragments into a manageable flow.
產品做了什麼What the product does
多通路租客溝通Multi-channel tenant communication
租客可以透過 LINE、Email 或 Web 聊天提問，系統保留對話脈絡，不讓資訊散在不同 app。Tenants can ask through LINE, email, or web chat while the system preserves conversation context instead of scattering it across apps.
自助預約看房Self-service viewing bookings
租客在公開物件頁面自助選時段，系統同步 Google Calendar，減少房東來回確認。Tenants book viewings from the public property page; the system syncs with Google Calendar and reduces back-and-forth confirmation.
房東 CLILandlord CLI
房東不用只靠後台 UI，可以用 Go / Cobra CLI 管理物件、對話、預約與 Q&A 知識庫。Landlords are not limited to an admin UI; a Go / Cobra CLI manages properties, conversations, bookings, and the Q&A knowledge base.
AI / Agent 整合AI / Agent integrations
AI Provider 可插拔Pluggable AI provider
定義 AIProvider 介面，OpenAI / MiniMax 可透過環境變數切換，避免把產品綁死在單一模型。Defines a pluggable AIProvider interface so OpenAI / MiniMax can be switched through environment variables instead of locking the product to one model.
AI 自動回覆 + 轉交AI auto-reply with human handoff
AI 根據物件資料與 Q&A 知識庫回答租客；回答不了時轉交房東，房東回覆再變成新的 Q&A 建議。AI answers tenants using property data and the Q&A knowledge base; when it cannot answer, it hands off to the landlord and turns the reply into a new Q&A suggestion.
CLI Agent 擴展架構CLI agent extension architecture
landlord-assistant spec 裡已設計 CLI natural language query、tool calling 與 Agent Core，讓房東未來能用自然語言查系統狀態。The landlord-assistant spec includes CLI natural-language queries, tool calling, and an Agent Core so landlords can query system status in natural language.
品質不是口號Quality signals
不是 toy chatbotNot a toy chatbot
AI 被接到房源、預約、Q&A、LINE webhook 與 Google Calendar，而不是停在單純聊天。AI is connected to properties, bookings, Q&A, LINE webhooks, and Google Calendar rather than stopping at a toy chatbot.
房東與租客雙邊流程Both sides of the workflow
同時考慮公開物件頁、租客預約、房東管理、媒體上傳與對話記錄。It covers both sides of the workflow: public listing page, tenant booking, landlord management, media upload, and conversation records.
部署與資料邊界清楚Clear deployment and data boundaries
Next.js / Vercel 處理 Web + API，Supabase 處理 Postgres / Auth / Storage / RLS，CLI 透過 HTTP 呼叫 API。Boundaries are clear: Next.js / Vercel for web + API, Supabase for Postgres / Auth / Storage / RLS, and the CLI calls APIs over HTTP.