跳到主要内容

分包处理方

本文件为英文版。中文译本将在法律审阅后提供。

Last updated: 31 October 2026

This list will be reviewed by counsel; we will tell you before any change.

BOOK IN CHAT PTE. LTD. (UEN 202646792Z) (BookInChat) uses the service providers below to run BookInChat. They may receive personal data of Shops' users and Customers. Each is bound by written terms requiring protection of personal data (for the AI providers, their published API terms and data processing terms). Where a provider processes data outside Singapore (or, for Malaysian Customers, outside Malaysia), see the Privacy Policy, section 7.

AI models

BookInChat uses one or more of the AI model providers below, reached only through BookInChat's own AI gateway (LiteLLM), for all Shops, including clinics. They receive the conversation text and context needed to understand a message and draft or send a reply, including health details at clinics (after the patient agrees to the clinic's notice), and, depending on the models in use, voice notes to transcribe and images to read. Images from clinic patients are never sent to an AI model. Data may be processed outside Singapore and Malaysia.

Since 17 October 2026, NUS SoC LaaS and Google (Gemini API) are our main AI providers. AI tasks include checking messages and replies, understanding and sorting messages, helping with bookings, writing replies and writing background summaries. Each AI task runs on NUS or Google Gemini, whichever is better tested for that task; where configured, the other provides a backup. For some Shops, OpenAI (OpenAI API) drafts replies and helps with bookings first. Anthropic (Claude API) is used for a task only once it has tested better for that task. Google transcribes voice notes first (sent as audio), with NUS SoC LaaS as the backup. Text sent to Google has names and other details masked where applicable. When a Shop has the AI suggest drafts, its staff may see two drafts side by side, one from each provider, without being told which is which.

Per task today (the provider used first; where configured, the other is the backup):

  • Checking messages for urgent or medical content and sorting them: NUS SoC LaaS.
  • Screening messages for attempts to misuse the assistant: NUS SoC LaaS.
  • Helping with bookings: NUS SoC LaaS; for some Shops OpenAI (OpenAI API) first.
  • Writing replies: Google (Gemini API); backup NUS SoC LaaS; for some Shops OpenAI (OpenAI API) first.
  • Checking replies before they are sent: NUS SoC LaaS.
  • Transcribing voice notes: Google (Gemini API); backup NUS SoC LaaS.
  • Reading images (never at clinics): NUS SoC LaaS.
  • Background summaries and search: NUS SoC LaaS.
  • Anthropic (Claude API): not yet used for any task.

Where each provider processes data is given in its entry below. How long a provider keeps the data it receives is set by its terms (Google, paid Gemini API: a limited period, only to detect abuse).

  • National University of Singapore, School of Computing: LaaS (NUS SoC LaaS, AI model service). Role: understanding and sorting messages, checking messages and replies, helping with bookings, drafting replies, background summaries, transcribing voice notes (as the backup to Google) and reading images: first choice for the tasks where it tested better, and a backup to Google where configured; and search. Location: as configured; may include regions outside Singapore and Malaysia.
  • Alibaba Cloud (Singapore) Private Limited: Alibaba Cloud Model Studio (Qwen models). Role: understanding messages and drafting replies; transcription and image reading where those models are used. We use Alibaba Cloud's international service, not its Chinese mainland region. Location: Singapore region only. Alibaba Cloud's published terms say it does not use customer content to develop or improve its Model Studio models without separate consent.
  • Google Asia Pacific Pte. Ltd.: Gemini API (Google Gemini models, paid service). Role: understanding messages, checking messages and replies, helping with bookings, drafting replies and background summaries: first choice for the tasks where it tested better, and a backup to NUS SoC LaaS where configured; transcribing voice notes (first choice); image reading where those models are used. Location: any country where Google or its sub-processors have facilities; data may be stored briefly or cached there. Google's published terms say it does not use prompts or responses on the paid Gemini API to improve its products, and logs them for a limited period only to detect abuse.
  • OpenAI, L.L.C.: OpenAI API (OpenAI models, paid API). Role: for some Shops, drafting replies and helping with bookings first, with NUS SoC LaaS or Google as the backup. Location: United States. OpenAI's API terms say it does not use API data to train its models, and it keeps API data for up to 30 days, mainly to detect abuse.
  • Anthropic, PBC: Claude API (Anthropic Claude models, paid API). Role: only the AI tasks where it tested better in our comparisons, shown in "Per task today" above (none yet), with the providers above as the backup. Location: United States. Anthropic's commercial terms say it does not use API data to train its models, and it keeps API data for up to 30 days, longer only where the law or a safety investigation requires.

Which of these providers handles a given message can change with availability, cost and quality.

Hosting and storage

  • Unihost: server hosting for the application and database. Receives all stored data. Location: Singapore.
  • Cloudflare, Inc. (United States): DNS, network protection, tunnel, email routing, Turnstile bot checks and R2 file storage. Receives traffic in transit (IP addresses and request data), emails sent to our addresses, Turnstile signals, and files and backups stored in R2. Location: Cloudflare's global network; Cloudflare does not promise a storage region for R2.

Messaging channels

  • Kapso, Inc. (United States): our WhatsApp Business Platform connection for numbers connected through Kapso. Receives WhatsApp messages and media in transit, phone numbers, profile names and message status. Location: United States and Chile.
  • Meta Platforms (WhatsApp Business Platform, including the WhatsApp Cloud API we connect to directly for numbers not connected through Kapso): delivers WhatsApp messages and acts as our processor for them under Meta's WhatsApp Business data processing terms. Receives WhatsApp messages and media, phone numbers, profile names and metadata. Location: global. Meta also processes some of this data under its own terms.
  • Meta Platforms (Facebook Messenger, Instagram and Threads APIs): delivers messages between Customers and a Shop's Facebook Page or Instagram account, and the Shop's replies to comments on its own Threads posts. Receives the messages and replies, the account IDs Meta gives the Shop's Page or account, profile names and pictures, and metadata. Location: global. Meta also processes this data under its own terms.

Operations and billing

  • Plus Five Five, Inc., trading as Resend (United States): transactional email (sign-up, password reset, booking confirmations, invoices, notices). Receives the recipient's email address and name and the email content. Location: United States.
  • Functional Software, Inc., trading as Sentry (United States): error monitoring. Receives technical error reports. Location: United States or European Union, depending on the account region; some account data is always held in the United States.
  • HitPay Payment Solutions Pte. Ltd. (Singapore): subscription billing and payment processing. Receives the Shop owner's name, email, billing details and payment status; card data is handled by HitPay and not stored by BookInChat. Location: Singapore.

Website analytics

  • Google LLC (United States): Google Analytics, for our website visitors who accept analytics cookies only (never the staff app, Shop Customer Data or WhatsApp messages). Receives pages viewed, referring sites, demo link clicks with their source code, device and browser data and approximate location; keeps event-level data for 14 months. Location: the United States and other countries where Google has facilities. See the Cookie Policy.

Changes to this list

We will email Shop account owners at least 3 hours before a new or replacement sub-processor starts processing Shop Customer Data, and update this page, unless an emergency requires a faster change (in which case we will tell you as soon as practicable). Shops may object; a Shop that objects and leaves receives a pro-rata refund of unused prepaid time, as set out in the Data Processing and Shared Control Agreement, clause 10.2.

分包处理方 | BookInChat