Privacy

Privacy notice

Written to be checked rather than skimmed. Every item below is what the software actually does, taken from the code, not a generic template.

1. Who this covers

Two different groups of people, with different relationships to us:

2. What the widget collects, exactly

This is the complete list. It is short by design, and the reasoning is written into the source code rather than decided by a policy document.

WhatWhy
visitor tokenA random identifier we generate. It lets a returning visitor continue the same conversation instead of starting again. It is not linked to any identity you have not given us.
page URLWhich page the conversation started on, so the agent can be useful immediately.
referrerWhere you arrived from, if your browser sends it.
timezoneRead from your browser settings. Used so that any follow up arrives during your working hours, not ours.
journeyFor up to the last 25 pages: the path, the page title, how long you stayed, how far you scrolled, and a product identifier if the page declares one.
messagesWhat you type to the agent, and what it replies.
voice notesIf you speak instead of typing, the audio of what you say, only for as long as it takes to turn it into text. See section 3.
contact detailsName, email, phone, only if you choose to give them in the conversation.

What it deliberately does not collect

No clicks. No keystrokes. No mouse movement. No form contents from elsewhere on the page. A page you looked at for under 1.5 seconds is not recorded at all. The limit of 25 pages is enforced in code, and older entries are discarded. This is meant to be a hint about the last few minutes, not a profile.

The agent is never allowed to repeat what you browsed. Browsing history is used only to form a guess that the agent then confirms by asking you a normal question. It is forbidden from stating, hinting at, or quoting which pages you visited, and that rule is enforced by an automated check on every message rather than by an instruction. If you ask the agent how it knew something, it will not pretend it did not have context.

3. Voice notes, and why the audio is not kept

Some places you can talk to the agent let you speak instead of type: the microphone button in the widget, and voice notes on messaging platforms where the agent is connected. When you do, the recording is sent to a speech to text processor that turns it into written words, and those words are what the conversation actually uses.

The audio itself is not stored. It exists only for the seconds it takes to transcribe, and is discarded after that. What is kept is the text, treated exactly like a message you typed. In the widget, you see the transcribed words in the box and confirm them before anything is sent.

The speech to text processor is named in section 12.

4. Storage on your device

The widget uses your browser's local storage, not cookies. Two keys are written: ampa_visitor_token and ampa_journey. Clearing your browser's site data for that website removes both, and the next conversation starts fresh.

This section is about the widget, which runs on our clients' websites. What we measure on ampa.chat itself is a separate matter and is described in section 15.

5. Where the data goes

Conversations are processed on our servers. To produce a reply, the message and the relevant approved facts are sent to a third party model provider. Providers are listed in section 12 and may change; we will keep that section current.

Provider keys are never stored on a client's website. The only credential that reaches a client's server is a site key, which identifies which business the widget belongs to and authorises nothing else.

6. Being answered in your own language

When a client turns on the language feature, the agent can read and reply in languages other than English. To do that, messages may be passed to language processors that translate them. The translation is used to understand you and to answer you; the same truth rules apply in every language, so a translated answer can still only state facts the business has approved.

The language processors are named in section 12. If a client has not turned this feature on, no translation happens.

7. Conversations that arrive from other platforms

A business can connect Ampa to messaging platforms so that a message sent there reaches the same agent: Telegram, WhatsApp, and Facebook Messenger and Instagram through Meta. The business connects and enables these itself, using its own number and accounts.

When you message a business on one of these platforms, that platform handles the message first and under its own terms and privacy policy, which are separate from ours and outside our control. What reaches us is the message content and the identifier the platform gives us so the agent can reply to the right person. Voice notes on these platforms are transcribed as described in section 3.

8. Payments

If a business pays for Ampa online, the payment is handled entirely by our payment processor. Card details are entered on the processor's own secure checkout and are handled by them. They never touch Ampa's servers, and we never see or store a card number. What we receive back is confirmation that a payment succeeded, so the account can be activated. The payment processor is named in section 12.

9. How long we keep it

Being accurate rather than reassuring: we do not yet run an automated deletion schedule. Conversations are retained while the client's account is active. A defined retention period is something we are working on, and this section will be updated with the specific period when it is in place rather than before.

In the meantime, deletion on request is honoured. Email info@ampa.chat and we will remove a conversation, or everything associated with a visitor token, and confirm when it is done.

10. Your rights

Depending on where you live, you may have the right to ask for a copy of your data, to have it corrected or deleted, to object to how it is used, or to complain to a regulator. Email info@ampa.chat and we will act on it. If you contacted the agent on a client's website, we may need to pass the request to that business, and we will tell you if so.

11. Marketing and opting out

If a conversation continues by email, every message carries an unsubscribe link, and one click is enough. The agent also treats "stop", "remove me", "not interested" or any equivalent phrasing in a normal reply as an opt out. Opt outs are permanent and are checked before every single send.

12. Processors we use

13. Talking to an AI

The agent is an AI. If you ask it whether it is, it will tell you the truth. It is not permitted to deny it under any circumstances.

14. Contact

Questions, requests, or complaints: info@ampa.chat.

15. What we measure on this website

Everything above describes the agent and the widget, which run on our clients' sites. This section is about ampa.chat itself, the site you are reading now.

Before you choose, nothing is loaded

We use Google Analytics 4 to understand which pages bring people here and which of our own adverts work. It is not loaded when the page loads. The script that draws the consent banner is served from this domain, contacts nobody, and only fetches Google's tag after you press Allow. If you press No thanks, or close the page without answering, no analytics script is fetched at all and Google is never contacted. That is a deliberate choice: a common alternative sends Google a so-called cookieless ping before you answer, which still hands over your IP address, and we did not think that was defensible on a site whose whole argument is that it does not do things it cannot defend.

Your answer is remembered in this browser's local storage under ampa_consent. Clearing site data for ampa.chat removes it and you will be asked again.

If you press Allow

Google Analytics 4 sets its own cookies and records the pages you visit, roughly where in the world you are, and a small number of named actions: opening the chat, giving the agent an address, reaching the pricing section, clicking a plan, downloading the WordPress plugin, reading a documentation page, and reaching each screen of the signup at /start. Google Analytics 4 truncates IP addresses before storing them and offers no setting to turn that off. We do not send it your name, your email address, or anything you type into the chat.

Where you came from

If you arrive from an advert or a tagged link, the campaign parameters in the address (utm_source and its companions, or an advertising click identifier such as gclid) are carried through to the signup form and, if you create an account, stored against that account so we know which of our own efforts brought you. This happens whether or not you allowed cookies, because it is not stored on your device and it describes our marketing rather than you. If you do not create an account, nothing is kept.

Signing up

Creating an account at /start stores what you type, which is your business name, your website, your email address and a password we only ever keep as a hash. That is a service record, not analytics, and it is not covered by the banner.

Session recording

We do not currently use any session replay or heatmap tool on this site. If that changes, this section will name the tool before it is switched on, and it will sit behind the same Allow.

Last updated 7 September 2026.