The short version: your data is yours, it stays yours, and nobody else gets to rummage through it. The long version follows, because privacy deserves more than a slogan.

#What It Does

ambientChat takes a fundamentally simple position on privacy: the data you put into the app belongs to you, is accessible only to you, and exists to serve you. There's no advertising model, no data marketplace, and no shadowy third party getting a copy of your grocery receipts.

Your inventory, documents, chat history, and location data are stored in a secure cloud account protected by Firebase Authentication and Firestore security rules. These rules enforce that each user can only read and write their own data — not just as a policy, but as a technical constraint enforced at the database level. Even if someone managed to guess your document IDs, the database itself would refuse to hand over the data. It's like having a bank vault that checks your ID at the molecular level.

When you talk to the AI, your context (inventory, location, conversation history) is sent to whichever AI provider you've selected — OpenAI, Google Gemini, or Anthropic Claude — to generate a response. This is necessary for the AI to be useful, but the data is sent per-request and is subject to each provider's data handling policies. You control which AI provider you use and what context is shared.

#Location Precision

When the AI helps you with something location-aware — identifying a monument in a photo, answering "where did I leave my keys?" — it uses your location. You decide how precise that location is, separately for each kind of AI call:

  • 📍 Exact GPS (default) — coordinates like 40.4168°N 3.7038°W plus city and country. Best accuracy; the AI can identify a specific landmark, recommend the closest restaurant, or distinguish two similar places. Trade-off: your precise coordinates leave your phone and reach the AI provider.
  • 🏙️ City — city + country (e.g. "Madrid, Spain"). Good balance for most queries — enough for "what monument is this?" without precise coordinates.
  • 🌍 Country — country only (e.g. "Spain"). Minimal signal; some location-dependent answers will be less accurate.
  • 🚫 Off — no location data sent. Maximum privacy; the AI works blind.

The two independent settings:

Setting When it's used
Photos (scan) When the AI looks at a photo or document you scanned
Chat When the AI answers a question referencing your inventory, documents, or context

Web searches never send your location: the AI searches with your words only, so there is no setting for them.

#Adjusting your precision

  • iOS: Control → Privacy policy → Location precision. Tap any row, pick from the four options.
  • Web: Control → Location precision card. Same two rows.
  • Just ask: tell the AI "Send less location info in chat" or "Use my exact GPS for photo scanning." It can adjust the settings on your behalf and will surface a notification with an undo whenever it does.

#Defaults

New accounts start at Exact GPS for both — maximum AI utility out of the box. Dial back any time. If the AI changes a setting on your behalf (because the next query would benefit), you'll see a notification with an Undo button.

#How to Manage It — Ask the AI

The fastest way to review or change any privacy setting is to ask the AI in chat:

  • "Guide me through the privacy controls for location." → The AI reads your current location-precision settings and offers to change them.
  • "What can I control about my privacy?" → The AI lists every privacy-related setting along with what it does.
  • "Send less location info when scanning." → The AI changes your scan-precision setting on the spot (with a notification + undo).
  • "Who has access to my data?" → The AI reads your third-party context-sharing settings.
  • "Hide adult beacons." → The AI updates your beacon-category notification preferences.

This works in regular chat and in voice/Conversation Mode — you don't need to leave the conversation to manage settings. The AI uses MCP tools to read your current state and apply changes, and surfaces every change as a notification with an Undo button so nothing happens behind your back.

#Settings the AI can manage for you

Setting What it controls Default
Location precision (scan / chat — independent) What the AI sends about your location — exact GPS, city, country, or off Exact GPS
Voice preferences TTS voice name, playback speed, auto-speak Sage, 1.0×, off
Beacon notification categories Which beacon categories (retail, hospitality, adult, etc.) trigger notifications All on
Context sharing Whether outside AI apps may read your context at all (the master switch) Off
Each connected AI app Pause or disconnect one app — Claude Desktop, ChatGPT, Grok, Muse — without touching the others On when you connect it
Profile Display name, nickname, profile photo, home and work addresses Empty
Security (Control → Profile information on iOS; Control → Security on the web) Email yourself a reset link, or change your password in place — type your current one, then the new one twice. Shown on any account that can sign in with a password, including an Apple or Google account that added one —

#The GUI is still there if you prefer it

Every setting has a screen too — useful if you want to scan the full list visually.

  • iOS: Open the Control tab and choose the relevant row (Privacy policy, Voice, Application behavior, Profile information)
  • Web: /control — all settings live on one long scrollable page with anchor links

#Which AI apps can read your context

Every AI app you connect to ambientChat — Claude Desktop, Claude Code, ChatGPT, Grok, Meta's Muse or any other — appears in Context Sharing on its own row, as soon as you approve it. One row per app: if you use Claude Desktop on a laptop and a desktop, both are the one "Claude Desktop" row, with "2 devices" under it.

Each row shows when the app last read your context and what it read recently, and has two controls:

  • The switch pauses it. A paused app stays connected, but every request it makes is refused — the others keep working. Turn it back on any time.
  • Disconnect signs it out on every device. It can't read anything until you connect it again from that app and approve it.

Either takes effect on the app's next request, within about a minute. The master Context Sharing switch above the list still stops them all at once.

Access alerts. When an outside AI app creates, edits, reads, moves or deletes one of your documents, your iPhone tells you what it did, to which document and where, right after it happens — e.g. "Grok created 'Website Brainstorm' in Product › Marketing. Use Chat or Control to stop." The first time an app touches a document you're alerted; repeats of the same thing to the same document stay quiet for 15 minutes. Anything that isn't about one document ("Grok accessed your Inventory") alerts at most once a minute per app. To stop an app, ask in Chat ("stop Grok's access"). These alerts are time-sensitive, so they come through Driving Focus and similar Focus modes, and reach your Watch when it mirrors your phone. You can allow or block them, and their time-sensitive delivery, for ambientChat in iOS Settings → Notifications.

Tap an alert to see it in Context Sharing. It opens at the app that did it, ready to pause or disconnect. Context Sharing repeats the alert's own words — on the phone in the Access Log and each app's Last: line, on the web in each app's list of recent calls — "Grok created 'Website Brainstorm' in Product › Marketing." — with the names shown even if you hide them on the lock screen. Calls from before this change, and calls that failed, show only what was used ("Documents").

Turn alerts off, or keep them but hide names. In Context Sharing (phone or web), Notify when accessed turns these alerts off for every app. Hide document names in alerts keeps them but leaves out the document and folder, so your lock screen says only "Grok created a document." Or just ask: "stop the access alerts" or "don't show document names in alerts."

Check that alerts reach you. Tap Send me a test alert in Context Sharing (phone or web), or ask the assistant to "send me a test alert". It sends the same alert to every device registered on your account and tells you, device by device, whether Apple accepted it. If Apple accepted it and nothing appeared, the problem is on that device. A registration that is out of date is removed.

In chat: "Which AI apps can see my data?", "Pause ChatGPT", "Disconnect Muse". The assistant asks before changing anything. An outside AI app can pause or disconnect apps when you ask it to, but only you can turn one back on — from the app's settings or ambientChat's own chat.

  • iPhone: Control → Context Sharing (in the Sharing group, or search Control for it). You can also ask in chat — "Which AI apps can see my data?" — or tap an access alert, which opens Context Sharing at that app
  • Web: Control → Context Sharing

When an AI app wants to show you a screen. Claude, ChatGPT or another connected app can send a screen to you in ambientChat. It arrives as a card — "Claude wants to show you Documents", Nothing changes until you choose. — with Open and Not now. Nothing moves until you tap Open, and an app you have switched off in Context Sharing can't send one. The card appears in the ambientChat app on your phone — or, if you're using ambientChat in a web browser right then and the web has that screen, in the browser. A screen that only the phone has always goes to your phone.

#Exporting your data

Independent of the privacy settings above, you can download everything ambientChat knows about you. See the Data Export & Import guide for details. In chat: "Export my data."

#Deleting your account

Ask the AI: "Delete my account." The AI runs the pre-flight check, summarizes what will be removed, and creates a pending confirmation that you confirm explicitly. Or delete it yourself: in the app, Control → My Subscription → Delete account and all data; on the web, Control → Account → Delete Account.

There is no "undo" for account deletion, and it deletes any export we're holding for you too. If you might want your data later, export it and save the .zip to your own device first. The Delete Account screen tells you whether you already have an export (and when it was made), or that one is still being made, and its Export or download your data button takes you to Your Data Export. In the app, going back from there returns you to Delete Account. Exporting is optional: you can delete without it.

Deleting your account also disconnects every AI app you connected (Claude, ChatGPT and the rest have to sign in again, and can't on an account that no longer exists), removes your Desktop AI Macs from the account, and removes the records our service keeps about your use of it: which tools ran with what you asked, your searches, your sign-ins, and the AI descriptions and narrations of your own photos.

If one of the last cleanup steps needs another try, you'll see "Account deletion is finishing; it completes automatically." ambientChat finishes it by itself, and there's nothing you need to do.

#MCP Tools Reference

These privacy and settings tools are available via Claude Desktop and other MCP clients:

  • get_privacy_preferences — Location precision per AI call type (change it with update_user_setting, key privacy)
  • get_context_sharing — The master context-sharing switch and the access-alert switches (change with update_user_setting, key context_sharing)
  • list_connected_apps — Every AI app connected to your account, its switch, and what it read recently
  • set_connected_app_sharing — Pause or resume one app
  • disconnect_connected_app — Disconnect one app on every device
  • get_notification_preferences — Beacon category notifications (change with update_user_setting, key notifications)
  • get_account_deletion_precheck — Pre-flight check before deletion
  • request_account_deletion — Request deletion (creates a 5-minute confirmation)
  • export_user_data — Export all your data before deletion

For the full tool reference, see MCP Tools — Complete Reference.

#Tips & Tricks

  • Just ask the AI. Anything below — "what's my voice speed?", "who can see my inventory?", "stop the AI from using my exact GPS" — is a one-sentence request away. Voice users especially: you should never have to look at a screen to manage these.
  • The AI may adjust ambient settings on its own (with notify + undo). For example, when scanning a monument it may temporarily upgrade your location precision to exact GPS so the identification is accurate. You'll see a notification with an Undo button after the response. Nothing happens silently.
  • Per-AI-call independence. "Location precision" isn't one switch — it's two: scan and chat are independent. You can be precise for landmark identification and coarse for everything else.
  • Export before delete. Account deletion is a one-way door — download your data first if you might want it later.

#What Data Is Collected

Here's a transparent breakdown:

Data Type Stored Where Who Can Access
Inventory items Firebase (your account) Only you
Documents (receipts, manuals, etc.) Firebase (your account) Only you
Chat history Firebase (your account) Only you
Location trails (if enabled) Firebase (your account) Only you
Last known location (phone) Firebase (your account) Only you. Saved when you send a chat message; if the phone's last reading is over 10 minutes old, the app asks for a fresh one while it is open (never more than once every 10 minutes, and never asking for permission). Saves and scans use it to record where an item was
Beacon visit history Firebase (your account) Only you
AI provider interactions Sent per-request to selected provider You + AI provider (per their policies)
Usage metrics Aggregate analytics anonymized, not personally identifiable
Account info (email, auth) Firebase Authentication Firebase infrastructure

For the full, named list of every outside service that may process your data — what each one does and where it's based — see our Subprocessors page.

#What Is NOT Collected

  • Your data is not sold to third parties. Ever.
  • There is no ad targeting based on your inventory or location.
  • Your specific items, documents, and conversations are not used to train AI models by ambientChat. (Individual AI providers have their own training data policies — review them if this matters to you.)
  • No third party gets your data by default. Context sharing is strictly opt-in.
  • A place you tap is not saved as where you are. With chat location Off, tapping Ask AI on a place in Radar, or a tour narrating the zone you walk into, still tells the AI that place — you asked — for that one answer. It is never saved as where you are, and later answers in that chat are not told it again (your question and the AI's answer stay in the conversation, as everything you say does).

#Known Limitations

  • AI providers receive your context per-request. When you ask a question, the selected AI model (OpenAI, Gemini, or Claude) receives the context needed to answer. This is how AI works — it needs your data to help you. Each provider has its own data retention and training policies, which are outside ambientChat's control.
  • Aggregate analytics are collected for service improvement. Things like "how many users scanned a barcode today" or "average response time for chat queries." These are anonymized and don't identify you personally, but they do exist.
  • At chat location Off, Radar answers are not prepared ahead. With Off, the app does not ask the AI about the exhibits around you before you tap one, or prepare their spoken Listen narration — that would tell it where you are. So a card's AI answer or Listen takes a moment to start at Off, instead of coming at once. A guided tour you turned on still prepares the zone you walk into.
  • Account deletion is irreversible. Once deleted, your data cannot be recovered. There is no 30-day grace period, no recycle bin, no "are you really sure?" email three weeks later. It's gone.
  • Firebase security rules are the enforcement layer. Privacy isn't just a promise — it's enforced by database rules that physically prevent cross-user data access. But like any system, it depends on correct configuration. Security rules are regularly reviewed and tested.

#Optional website measurement

On the redesigned public website, Website measurement in the footer lets you allow or decline optional first-party measurement. Before permission, the page creates no visitor identifier and sends no acquisition events. We respect Global Privacy Control and Do Not Track signals.

If you allow it, a random browser credential lasts up to 30 days. If you sign in or use web checkout, the visit can be linked to confirmed purchases and later refunds or subscription changes for the analytics retention period, until you turn measurement off. This measurement does not collect prompts, documents, inventory or email addresses.

Turning it off immediately stops collection on that browser and requests removal of its account attribution link. If offline, the site keeps only the credential needed to finish that withdrawal and retries when the connection returns; a notice stays visible until the unlink is confirmed. Existing anonymous aggregates and necessary billing records follow the retention policy. Your decision does not subscribe you to either newsletter.

#Version History

Version Date What Changed
30 2026-11-14 Explained optional website measurement, account attribution, privacy signals and durable withdrawal. (AMB-3292)
29 2026-11-11 On the iPhone, Context Sharing (which AI apps can read your context) now has its own row in Control, in the Sharing group, and turns up in Control's search — the same name and place as on the web AMB-3206
28 2026-11-07 A deletion with a cleanup step left over now says "Account deletion is finishing; it completes automatically." on the phone, the web and in chat, and ambientChat finishes it by itself AMB-2895
27 2026-10-27 Steps now name the tabs the app has today — Context → Docs and Control — instead of the retired "My Things" and "Account" AMB-2584
26 2026-10-27 The card an AI app sends can appear in the browser you are using, as well as on your phone. Phone-only screens still go to the phone AMB-2584
25 2026-10-27 On the iPhone, the way to the list of AI apps with access is chat, or tapping an access alert. The guide named a Control row the phone does not have AMB-2584
24 2026-10-27 Explains the card an AI app can send to your phone ("… wants to show you …", Open / Not now): nothing moves until you tap Open AMB-2584
23 2026-10-24 At Off, the Radar prepares no Listen narration before you tap. With chat location Off, the Radar no longer prepares the spoken narration of nearby exhibits in advance, since that would tell the AI where you are. Listen now takes a moment to start. A guided tour you turned on still prepares the zone you enter. AMB-2481
22 2026-10-24 At Off, Radar asks the AI nothing before you tap. With chat location Off, the Radar no longer prepares AI answers for the exhibits around you in advance, since asking would tell the AI where you are. A tapped answer now takes a moment to write instead of appearing instantly. AMB-2478
21 2026-10-24 Your last known location is listed. The data table now shows that the phone saves your last known location when you send a chat message, refreshing it if it is over 10 minutes old, so an item you save while sitting still is placed where you are. AMB-2452
20 2026-10-24 A place you tap is named for that answer only. At chat location Off, tapping Ask AI on a place in Radar, or a tour narrating the zone you enter, still tells the AI that place — you asked — for that one answer. It is never saved as where you are, and later answers in that chat are not told it again. At City or Exact nothing changes: follow-up answers still know the place and its details. AMB-2446
19 2026-10-24 The web-search location setting is gone. Web searches never sent your location, so the switch changed nothing; Location precision now has two settings, scan and chat, on the phone and the web. AMB-2426
18 2026-10-18 Tapping an AI-access alert opens Context Sharing at the app that did it, and the access list there now says what the alert said ("Grok created 'Website Brainstorm' in Product › Marketing."). (AMB-2115)
17 2026-10-18 The "Notify when accessed" switch now saves (it used to flip back), and a new "Hide document names in alerts" switch keeps the alerts but leaves out the document and folder — on phone, web and in Chat. (AMB-1994, AMB-2116)
16 2026-10-16 AI-access alerts now say what the app did, to which document and in which folder — sent after it happens, once per document, ending "Use Chat or Settings to stop." (AMB-2055)
15 2026-10-13 Context Sharing has a Send me a test alert button (phone and web): it sends the AI-access alert to each of your devices and shows whether Apple accepted it on each (AMB-2056)
14 2026-10-12 AI-access alerts ("Grok accessed your Lists") are now time-sensitive, so Driving Focus no longer holds them back (AMB-1993)
13 2026-11-07 A deletion still finishing says so. If a last cleanup step needs another try, the app, the web and chat say "Account deletion is finishing; it completes automatically." instead of claiming it's all gone (AMB-2895)
12 2026-10-22 Deleting your account reaches everything. It now also disconnects your AI apps, removes your Desktop AI Macs, and removes the service records about your use: tool calls, searches, sign-ins, and descriptions of your photos (AMB-2376)
11 2026-10-20 Change your password on the phone. The Security card on iOS now changes your password in place, as the web always has — no reset email needed if you know your current one (AMB-2207)
8 2026-10-16 The Settings tab is now called Control, on the phone and the web; the web address is /control (old /settings links still work) (AMB-2120)
1 2026-03-01 Initial guide
2 2026-05-19 Rewrote "How to Use It" → "Ask the AI" tool-first framing. Replaced stale screen breadcrumbs that referenced non-existent paths. Added the Settings table reflecting the actual MCP-controllable settings post-AMB-176 / AMB-224. Removed the "AI Provider = Gemini" default claim (that's no longer accurate — the user picks per conversation).
3 2026-06-07 Terminology: settings screen labeled "Web" not "Admin web" (the web interface)
4 2026-07-19 Added a link to the public Subprocessors page (the named list of service providers that may process your data).
7 2026-09-30 Every connected AI app now has its own row in Context Sharing — pause or disconnect one without touching the others, see what each read recently (iOS, web and in chat). Replaces the single "Claude" switch that turned off every app (AMB-1615)
6 2026-09-23 Deleting your account: the Delete Account screen (app and web) now links to Your Data Export and shows your latest export; exporting is encouraged, never required, and the exports we hold are deleted with the account (AMB-1360)