Privacy Policy

Last updated: July 2026

This Privacy Policy explains how Privas Intelligence Ltd, a company registered in England and Wales under company number 17195166 and trading as Privas AI, handles personal data.

We have written this policy to describe what our software actually does. Where a capability is not yet built, we say so rather than implying otherwise.

1. Who this policy is for

Privas AI handles two separate groups of people, and our legal role differs for each. Read the part that applies to you.

You are	Read	Our role
A business or individual with a Privas AI account	Part A	We are the controller of your data
Someone who chatted with an AI assistant on another company's website	Part B	We are a processor. That company is the controller

If you chatted with an assistant on a company's website and want your data deleted, that company decides — see section B.9 for how to reach them and how we help.

Part A — If you have a Privas AI account

A.1 What we collect

Account information. Your name and email address. Your email is held both by our authentication provider, Clerk, and in our own database, where it is refreshed from Clerk each time you sign in and shared with Stripe for billing.

Billing information. Your subscription plan, billing period, usage counters, and Stripe customer and subscription identifiers. We never receive or store your card details — those are collected and held by Stripe.

Content you upload. Documents, web pages, and other knowledge sources you add to train your assistant, plus images such as assistant avatars and domain icons.

Configuration you provide. This may include credentials for services you connect: your own mail server password, and authentication tokens for external tool servers you register. These are encrypted before storage (see section A.5).

Team and access data. Collaborator roles, and the email addresses of people you invite — including invitations that are never accepted.

Enquiry data. If you join a waiting list or submit a contact or demo request, we keep the name, email, phone number, company, and any notes you provide.

A.2 Why we use it

To operate your account and your assistants; to process payments and enforce plan limits; to send service messages such as usage warnings and billing notifications; to provide support; to keep the platform secure; and to meet legal obligations.

Our lawful bases are performance of a contract for operating the service and billing, legitimate interests for security, abuse prevention and service communications, and consent where you have given it, such as joining a waiting list.

A.3 How long we keep it

Account and content data is kept for as long as your account is open. When you close your account, please read section A.6 — the position today is narrower than you might expect.

Billing records are kept for seven years after the transaction to meet UK accounting and tax requirements.

Waiting list, enquiry, and unaccepted invitation records are currently kept indefinitely. We are introducing a retention limit for these; until then you can have yours removed at any time by emailing us.

The providers listed on our Sub-processors page, which forms part of this policy. In summary: Clerk for sign-in, Stripe for payments, Uploadcare for images, Spacemail for outbound email, Contabo for hosting in Singapore, Google Analytics for website usage, and Anthropic and OpenAI for processing the knowledge sources you upload.

We do not sell your personal data, and we do not share it for advertising.

A.5 How we protect it

Credentials you entrust to us — mail server passwords, external tool tokens, and the database credentials we generate for your assistant — are encrypted with AES-GCM using a unique nonce per operation before being written to storage. The encryption component validates its key at start-up and refuses to run if the key is invalid.

Each of your domains gets its own separate database for its knowledge and conversations, rather than sharing storage with other customers. Administrative access requires authentication and a role check on the specific domain.

Section B.6 sets out the limits of these measures honestly, including where our protection is weaker than it may sound. Those limits apply to your data too.

A.6 Deleting your account — the current position

You can delete your account from your dashboard settings. Doing so removes your Clerk identity and your records in our main database.

However, deleting your account does not currently delete the conversation databases belonging to your domains. Because of a defect in how account deletion cascades, those databases are left behind — disconnected from the application and no longer reachable through it, but not erased.

Deleting an individual domain does work correctly and destroys its conversation database, with one exception: the separate database holding cross-session visitor memory (see section B.4) is not currently removed by any automated path.

We are fixing both. Until then, email [email protected] and we will erase the remaining data manually and confirm when it is done. We would rather tell you this than let you assume a deletion happened that did not.

A.7 Your rights

If you are in the UK or EEA you have the right to access, correct, delete, restrict, or object to our processing of your data, and to receive it in a portable format. If you are in Indonesia, equivalent rights apply under Law No. 27 of 2022.

Email [email protected]. We will respond within 30 days. You may also complain to the UK Information Commissioner's Office at ico.org.uk, or to your local supervisory authority.

Part B — If you chatted with an assistant on a website

When you chat with an assistant powered by Privas AI on another organisation's website, that organisation decides what is collected and why. They are the controller. We store and process the conversation on their behalf, under a contract with them.

This part tells you what we hold, because you are entitled to know even though your rights are exercised against them.

B.1 What is stored when you chat

Each exchange is stored as a record containing:

Your message, exactly as you typed it
The assistant's reply
A rewritten version of your question, which the AI produces internally to search the knowledge base more effectively
Which passages of the organisation's knowledge base were used to answer
Results returned by any external tool the assistant called — for example a calculated insurance premium or an exchange rate
Technical information about why the assistant behaved as it did

If you provide your email address — because the assistant asked for it, because you requested a follow-up, or because you asked for the conversation by email — it is stored against your chat session, against a contact record, and in our outgoing email queue, where it remains alongside the message body.

Please treat a chat window as you would any web form: do not enter passwords, payment card numbers, national identity numbers, or health information.

B.2 What we do not collect

This is worth stating plainly, because it is unusual:

We do not record your IP address.
We do not collect geolocation data.
We do not record your browser's user-agent string, and we do not fingerprint your device.

None of these are collected anywhere in the visitor-facing product. This is a deliberate design choice, not an oversight.

B.3 What is stored in your browser

We do not use tracking cookies in the chat widget. We use your browser's local storage:

A session identifier — a random value that groups your messages into one conversation. It expires after 12 hours.
A visitor identifier — a random value used to count returning visitors in aggregate analytics. It has no expiry.
A copy of your conversation, so the chat survives a page refresh.

These are random values. They are not linked to your name or identity unless you provide it. Clearing your browser's site data removes them.

We are adding a consent step before the visitor identifier is set. Today it is set without asking, which we do not consider good enough.

B.4 Memory across conversations

This feature is currently active.

The assistant can remember information across your separate visits. When you chat, an AI model reads your messages and extracts entities — things you mentioned, such as a product, a company, or a place — and preferences it infers from what you said. These are stored in a memory database belonging to the organisation whose assistant you used, linked to your browser's session identifier, and may be used to give more relevant answers when you return.

Staff at that organisation can view what has been remembered for a session, and the reasoning the assistant used, through their dashboard.

Two things you should know:

The memory database is shared by all visitors to that organisation's assistant. Your data is kept separate from other visitors' by a filter applied in our software, not by giving each person their own database.
Memory is keyed to your browser. Clearing your browser's local storage breaks the link.

B.5 Analytics

We record a small set of events so the organisation can see how their assistant performs — for example that a chat was opened, that a question was answered, or that a link was clicked. We store the event name, the page address, and a few non-identifying parameters.

Before storage, parameters are stripped of anything resembling personal data. Both field names and values are checked, so an email address or phone number is removed even if it arrives under an innocuous label.

Analytics events are deleted after approximately 90 days. Because they are removed in monthly blocks, an individual event may persist for up to about four months. Aggregate daily totals, which contain no session or visitor identifier, are kept permanently.

Note also that because the chat window is served from privas.ai, Google Analytics currently loads within it, and Google receives the fact that a chat surface was opened. Google does not receive your messages. We are removing this from the chat surfaces.

B.6 How your conversation is protected — and the limits

What protects it. Each organisation's conversations live in a physically separate database with its own credentials, so one customer's assistant cannot reach another's data. Staff access requires authentication and a role check. Stored credentials are encrypted.

Where the protection is weaker than it sounds. We would rather you knew this than discovered it:

The endpoints the chat widget uses cannot require a login, because you are an anonymous visitor. They are restricted by checking which website the request came from — a check a normal browser enforces but which a determined party using a scripted tool can imitate. In practice your conversation is protected by the fact that its session identifier is a long random value nobody can guess, rather than by a password.
Within an organisation's memory database, separation between visitors is enforced by our software rather than by the database itself.

Do not send anything through a chat widget — ours or anyone else's — that you would not put in an ordinary web form.

B.7 Where your conversation goes

Your message is stored on our servers in Singapore.

To generate a reply, your message, the recent conversation, and the relevant parts of the organisation's knowledge base are sent to Anthropic in the United States. Your message is also sent to OpenAI in the United States to be converted into a numerical form used for searching the knowledge base.

If the organisation has connected an external tool, information the assistant extracts from your message may be sent to a server that organisation chose and controls. We do not control that server. Ask them about it.

If the assistant emails you, your address and the message pass through our email provider.

We do not train AI models on your conversations, and we do not sell or share your conversation for advertising.

B.8 How long your conversation is kept

Conversations are currently kept for as long as the organisation keeps their assistant. There is no automatic expiry today.

We are introducing a configurable retention limit so organisations can have conversations deleted after a defined period. Until that ships, we will not state a retention period we do not enforce.

Analytics events are removed on the schedule in section B.5.

B.9 Getting your data deleted

Start with the organisation whose assistant you used. They are the controller and the decision is theirs. Their privacy notice should tell you how to reach them; the assistant itself can usually tell you who operates it.

We will help. Contact us at [email protected] with the website and roughly when you chatted. We will identify the organisation, pass your request on, and act on their instruction.

Being straightforward about how this works today: there is no self-service button, for you or for them. Erasure is performed manually by our team on the controller's instruction. We are building a proper mechanism. In the meantime the outcome is the same — your data is deleted — but it takes a person, so please allow up to 30 days.

Part C — Applies to everyone

C.1 International transfers

Privas AI is registered in England and Wales, operates from Jakarta, Indonesia, and hosts its infrastructure in Singapore. AI processing takes place in the United States. Personal data therefore crosses borders as a normal part of the service.

For UK and EEA data we rely on the UK International Data Transfer Addendum or Standard Contractual Clauses, together with each provider's own transfer terms. For Indonesian data subjects we rely on the cross-border provisions of Law No. 27 of 2022 on Personal Data Protection.

C.2 AI models and your data

What we do. Privas AI does not train, fine-tune, or otherwise use customer content or visitor conversations to build machine-learning models. We operate no training pipeline. Our AI providers are called only to answer requests as they arrive.

What we can say about them. We use Anthropic and OpenAI through their standard published API terms. We have not negotiated a private agreement, a zero-retention arrangement, or an enterprise addendum with either of them, and we have no commitments from them beyond what they publish and apply to every API customer.

We mention this because many privacy policies state that "our AI providers do not train on your data" as though it were a promise the company had secured. In our case it would be a description of the providers' own published default, not something we have contracted for — and they can change their published terms without asking us. You can read their terms yourself; we will point you to the current versions if you ask.

One practical consequence you should know: under standard API terms a provider may hold submitted content for a limited period for safety and abuse-monitoring purposes, even where it does not use that content for training. We cannot switch that off.

The one place the platform learns from usage is a feature that records procedural patterns for a single organisation — which kind of tool tends to suit which kind of question, and which pieces of information are usually needed. It never stores the content of an answer or anything a visitor said, it is never shared between organisations, and a pattern is only retained once it has occurred across multiple independent sessions.

C.3 Children

Privas AI is a business product and is not directed at children. We do not knowingly collect data from anyone under 18. If you believe a child's data has reached us, email [email protected] and we will delete it.

C.4 Security incidents

If personal data we hold is compromised, we will notify affected customers without undue delay and within 72 hours of becoming aware, and regulators where required.

We are candid about the current limitation: our monitoring and alerting are not yet where they should be, so becoming aware may depend on a person noticing rather than an automated alert. Improving this is our highest-priority engineering work.

C.5 Changes to this policy

We will update the date at the top when this policy changes and, for material changes affecting visitor data, update the Sub-processors page beforehand. Customers with a Data Processing Agreement receive notice as set out in that agreement.

C.6 Contact

Privas Intelligence Ltd Email: [email protected] Website: https://privas.ai Registered in England and Wales, company number 17195166 Registered office: 71-75 Shelton Street, Covent Garden, London, WC2H 9JQ, United Kingdom Operations: Jakarta, Indonesia

We have not appointed a Data Protection Officer, as we are not required to. Privacy enquiries go to the address above.

About Privas AI

Privas AI is a privacy-first AI assistant platform designed to help organizations deploy conversational AI representatives on their websites.

The platform uses domain-scoped knowledge retrieval to ensure that AI responses are generated only from authorized knowledge sources.

Learn more about Privas AI:

Sub-processors: https://privas.ai/sub-processors
Terms of Service: https://privas.ai/terms
Cookie Policy: https://privas.ai/cookies
Documentation: https://privas.ai/documentation