AI assistant notice

What it is

The assistant is an AI, not the agent and not a person. Its replies are labelled AI assistant, and it is told to say it is an AI whenever it is asked. It answers questions about the agent, their listings and the off-plan projects on their page.

It appears only on an agent's own page (which can also be a workspace's front page, when that shows one agent), and only beside a number to reach them on WhatsApp (their WhatsApp number, or their phone number if they gave no WhatsApp number), so a person is always one tap away.

What it answers from

Only what the agent's page publishes: their public profile, listings and projects, a short profile the agent wrote or recorded and approved, and the agent's answers to common questions, which Rina writes in general terms and the agent can change; an answer with figures says the month they hold as of. It reads the page the same way the page does, so it sees nothing a visitor of the page could not see. It cannot see property owners or landlords, commissions, internal notes, ID documents or any private record.

It is told never to answer from general knowledge, never to invent or estimate a price, a fee or a term, and never to give legal, tax, visa, mortgage or investment advice. It can still get things wrong: check any detail with the agent before you rely on it.

The agent's profile and voice notes

An agent can record a voice note for their profile instead of typing it. ElevenLabs turns the recording into text. Rina deletes the recording from its storage as soon as it is transcribed, or could not be, and once ElevenLabs answers, even too late to use, asks it to delete its copy of the text, trying again if that fails. Anthropic drafts the profile from that text alone, and the agent checks and approves it.

The text is kept until the agent approves a profile, or 30 days after it was made, then deleted by a daily job; the agent can delete it sooner. Only the profile the agent approved goes with visitors' requests: never the recording, its text or a draft.

What not to type

Do not type an Emirates ID, passport or card number into the chat. The assistant cannot take payments or deposits, and cannot see or receive ID documents. Nothing removes what you type: it goes to Anthropic with each later request in the conversation, is kept until 30 days after the conversation's last message, and can be copied to the agent with your enquiry.

Reaching a person

A button to message the agent on WhatsApp is always at the top of the chat. When you ask for a person, or want to view, make an offer, or rent or buy, the assistant is told to show a button to continue with the agent on WhatsApp, and its fixed replies (when it cannot answer) show one too. The message these buttons prepare says only that you were chatting with the assistant. Your conversation is not sent to WhatsApp.

Questions it cannot answer

When nothing it has answers your question, the assistant passes the question to the agent as a notice in Rina, in its own words and without the rest of the conversation, and shows the button to message them on WhatsApp. To hear back, message the agent or leave your details. A few questions a day are passed on from one conversation, and from one network; past that, the assistant points you to WhatsApp. The question stays in the agent's notice with no end date yet, even after the conversation is deleted.

Passing your details to the agent

The assistant can pass your name and mobile number to the agent as an enquiry, with your email, what you are looking for, your budget and when you plan to move if you gave them. It is told to ask you first, and to pass on only what you gave it and agreed to share. Whatever you agree to, the latest part of the conversation (about the last 1,500 characters) goes with your details, including anything else you wrote in it. The assistant is not told to warn you of this.

It can pass your details on again later in the same conversation. With the same number, what was said since the last time is added to the same lead; with another number, it is an enquiry by that number. The limits on the page's enquiry form apply to it too, including a limit on enquiries from one number.

Rina keeps no separate record of your agreement, and no code checks it: the assistant judges it. Only the conversation shows it, and the conversation is deleted 30 days after its last message. The copy on the lead holds only its latest part, which may not include the agreement.

The enquiry becomes a lead in the agent's workspace. It goes to the agent whose page you were on. If your number is already a lead in that workspace, the enquiry is added to that lead and stays with the agent who has it, while they are still active. The conversation is copied into the lead's timeline and into a notice to that agent, whose title carries your name, and is usually emailed to them, through Resend, if Rina's email is set up and they have not turned email copies off. Repeat enquiries on a lead the agent already holds make at most one notice a day, and a member who has had 10 emails in 10 minutes gets no more until that time has passed.

The copy on the lead stays until the workspace deletes the lead, then for up to 30 more days. If the lead is merged into another, the copy moves to that lead and its agent, and stays until that lead is deleted. The copy in the agent's notice has no end date yet, even after the lead is deleted. An emailed copy stays in the agent's mailbox, where Rina cannot delete it.

What is sent to Anthropic

Anthropic's AI model writes the replies. Your browser never talks to Anthropic: Rina's server sends each request. It contains:

• the assistant's fixed instructions and the descriptions of what it can look up, the same for every page;
• the agent's name and first name and the workspace's name, as they were when the conversation started, sent as data marked as names only, never as instructions;
• everything you have written in this conversation, except a message the model declined to answer, and the assistant's earlier replies with the look-ups it made and any details it passed to the agent (your name, number, email, budget and when you plan to move, as it recorded them);
• the listings and projects the agent's page publishes, as the page shows them when your latest message arrives: their titles, places, sizes and published prices, the same for every visitor of that page;
• the agent's profile in their own words, only as the agent approved it, the same for every visitor of that page;
• the agent's answers to common questions, the same for every visitor of that page;
• the published details the assistant looked up to answer your latest message: the agent's public profile, listings and projects. Details looked up for an earlier message are not sent again.

Anything you type, such as your name, number or email, is part of your messages, so it goes too. Rina does not send Anthropic your network (IP) address, anything else about your device, or any private record.

[To be confirmed: Anthropic's terms for Rina's account: how long Anthropic keeps these requests, whether they can be used to train its models, and where they are processed.]

Who else handles it

• Convex runs Rina's database and server: it stores the conversations, and runs the code that calls Anthropic.
• Vercel hosts the chat's address on Rina's site, so every message you send, and every request the page makes for a reply, passes through it with your network (IP) address and your browser's details (the request's headers). Its bot check, Vercel BotID, checks each message first, and a message it cannot clear is refused. The check runs in your browser too, receives those headers, and may set cookies of its own.
• Resend sends the email copy of an enquiry, with the conversation and your name, to the agent.
• Anthropic writes the replies, as described above, and drafts an agent's profile from the text of their voice note.
• ElevenLabs turns an agent's voice note into text, when they record their profile. It receives the recording only, never a visitor's messages.

The privacy policy lists every service Rina uses.

What Rina keeps, and for how long

• Your messages and the assistant's replies, with the steps it took (its reasoning included) and the page details it looked up, are kept in Rina's database. A daily job deletes them once 30 days have passed since the conversation's last message.
• With the conversation, Rina keeps a keyed code made from the network (IP) address your latest message came from never the address itself, to hold each network to its share of the assistant (below). It is deleted with the conversation. The code is the same for your network wherever and whenever you use the assistant on Rina, so these records can be matched to each other, though not to your address without Rina's secret key. Rina keeps no browser details with a conversation, and its own code sets no cookie for the chat; Vercel's bot check may (above).
• The chat's key stays in the page's memory, and Rina's database keeps only a scrambled form of it (a hash). But the page sends the key in the address of each request it makes for a reply, every second or two while one is being made, so the key passes through Vercel. Anyone holding the key can read the conversation, and send messages in it, until it is deleted.
• Reloading the page starts a new chat. The earlier one is not deleted: it is kept for 30 days like any other, and you can no longer open it.
• Each message first passes Vercel's bot check, described in the privacy policy.
• What your network spent on the assistant in the workspace, and the record of each reply being made for it, under the same keyed code, is kept until the day or month it counts for is over, then deleted by a daily job.
• To limit misuse, Rina keeps the keyed code made from your network address, and the digits of any phone number passed on, in its records of repeated attempts. These have no end date yet.
• For each call to Anthropic, Rina keeps what it cost, how much text it used and how much of it came from the prompt cache, and which conversation it was for, with no conversation content. This has no end date yet.

Rina has no way yet to delete a conversation sooner.

[To be confirmed: whether Vercel's request logs keep the chat's key, which travels in the address of each request for a reply, and for how long.]

[To be confirmed: what Vercel BotID collects in the browser when the chat is used, whether it sets cookies or browser storage, and how long Vercel keeps what it receives (the privacy policy asks the same).]

[To be confirmed: how to ask for a conversation, or a copy of it attached to a lead or a notice, to be deleted, and who answers.]

Who sees it

The stored conversation is marked as private to the agent whose page it was on and to the workspace's owners and admins, but Rina has no screen that shows it to them or to anyone else in the workspace.

Once the assistant passes your details on, the agent who holds the lead, and the workspace's owners and admins, see the lead and the copy of the conversation on it. That agent may not be the one whose page you were on, if your number was already a lead in the workspace, and anyone the lead is later given to, or whose lead it is merged into, sees it too. Every member of the workspace can see which colleague holds a lead with your number, but not the lead itself.

[To be confirmed: who at Rina can read stored conversations, and how that access is recorded (the privacy policy asks the same for the whole database).]

When it is on

It is off unless a workspace's owner or admin turns it on, for all of that workspace's public agent pages. It can be on only with a plan that includes it (Team and Brokerage), on a paid plan or a partner plan, including while a payment is overdue, or on a workspace made before billing, which keeps every feature. The free trial does not include it, and a read-only workspace's pages do not offer it.

Rina pays for the assistant, up to a limit each month for each workspace: AED 5 per seat on Team and Brokerage (on a workspace made before billing, the same for each active member), unless Rina sets a different limit for that workspace. Rina spreads that limit over the month's days: each day the assistant answers until it has used that day's part, which is what is left of the month's limit shared over the days left, and then answers again the next day. One visitor's network can use about a share of a day's part (10%) and of the month's limit (25%): a message already being answered can finish past it. Once it has used its share, the assistant points that visitor to the agent's WhatsApp and goes on answering others.

When the month's limit is used, the assistant stops answering, and the page stops offering it, until the next month or until the limit is raised; visitors are pointed to the agent's WhatsApp, and the workspace's owners are told.

A message can be up to 1,000 characters, and a conversation can hold up to 30 of your messages. A conversation that grows too long for what is left of the month ends, and a new one can be started by reloading the page.

On test setups a test assistant gives scripted replies, labelled Test assistant, which quote the page details it looks up, and sends nothing to Anthropic. Rina's production build refuses to deploy while the test assistant is on.

More

How Rina handles personal data in general is in the privacy policy.

[To be confirmed: the legal basis for sending visitors' messages to Anthropic, and how to contact Rina about this notice.]