Oversee runs on your Mac, not in the cloud. We don’t sell your data, we don’t run ads, and the content of the things you watch stays on your Mac while a watch is running, with two optional exceptions. First, creating a watch can ask a cloud AI model to help draft its rule, which sends that region’s on-screen text and a screenshot of the window off your Mac. See AI-Assisted Rule Drafting below. Second, if you use the optional iOS companion, a notification’s text reaches your phone through our server and Apple’s push service, and a picture you ask your Mac for passes through our server on its way to your phone. Both are encrypted so neither we nor Apple can read them. See The iOS Companion below.
We only collect what we need to run the app, and you’re always in control.
Got questions? Email [email protected].
Who We Are
Oversee is built by Serial Labs Ltd, a small UK-based company.
We host our core server infrastructure, your account, licence, and error tracking, in the EU. The oversee.so website, including this page, is served through Cloudflare’s global network rather than from the EU alone. See Where Your Data Lives below for the full picture. We only work with vendors who meet strong privacy standards, and collect the bare minimum needed to run the app.
We do not process special-category data (such as health, ethnicity, religion, or political views) and do not knowingly collect such data from you.
The short version
The things you watch stay on your Mac while a watch is running. Screen regions, OCR text, the frames we analyse, the ML crops we save for training. We never upload any of that, apart from the two optional exceptions below.
The first exception: creating a watch can send that region’s on-screen text, a short description of it, and a screenshot of the window to a cloud AI model (Google Gemini) to help draft the rule, either straight from your Mac if you’ve supplied your own API key, or through our server if you’re signed in and haven’t. See AI-Assisted Rule Drafting below for exactly what’s sent and what, if anything, we keep.
The second exception only exists if you pair the iOS companion with your Mac: you can ask your Mac for a picture of what a watch is looking at. Your Mac takes it at that moment and encrypts it so only your paired iPhone can open it. It passes through our server, which can’t read it, holds it in memory for at most a minute until your phone collects it, and never writes it to disk. We can see that a request happened, not what it showed. If you also turn on pictures for finished-watch notifications, your Mac sends one the same way each time a watch finishes, without being asked. See The iOS Companion below.
The only things we keep on our servers are:
- Your account (email, name, auth provider)
- Your licence and which devices you’ve activated it on
- A small set of basic attributes from each device (app version, major macOS version, configured capabilities) so we can plan compatibility
- If you use the iOS companion, a device key and push token for your iPhone, so we can route notifications to it
- If you use the iOS companion, a small status record for your Mac while a watch is set to alert your phone: when it last checked in, and whether it went to sleep or shut down. That is how we can tell you if your Mac stops watching. The alert and the watch list inside it are encrypted to your phone, which we can’t read.
- Error reports, when they’re automatically submitted
- Debug bundles, only when you explicitly click “Send debug report” or flag a result
That’s it. Everything else lives on your device.
What We Collect (and Why)
We only collect data you give us directly, plus what’s automatically generated by using the app (licence checks, device telemetry, error reports). When you sign in with Apple or Google, we receive the small set of fields Apple or Google return to us.
Account
| Data | Purpose |
|---|---|
| Sign-in, transactional emails (magic-link codes, receipts), account recovery. Stored encrypted, alongside a one-way hash used only for lookup. | |
| Name | Personalises the app and support emails. Stored encrypted. |
| Apple / Google account info | Only if you sign in with Apple or Google. We receive your email, a stable provider user ID, and optionally your name. We do not request any other scopes. During sign-in, Apple’s or Google’s authorisation code briefly passes through our API (api.oversee.so/auth/apple/callback or .../auth/google/callback) to exchange it for a user profile. We do not retain the OAuth token after sign-in. |
| Marketing consent flag (opt-in) | Recorded only if you opted in via the Paddle checkout when you bought a licence. Default is off. We use this to decide whether you can receive promotional emails (announcements of new products, offers, etc.). |
| Product update preference | Whether you’ve opted out of operational product update emails (new versions, breaking changes, security advisories). Default is opted in for licensed users. Recorded the first time you click the unsubscribe link in one of those emails. |
Licence
| Data | Purpose |
|---|---|
| Paddle transaction and customer IDs | Link your purchase to your account so we can validate your licence |
| Licence activations | A record of which devices have been activated against your licence (up to 3 devices). Visible to you in Settings → Devices, where you can also deactivate a device. |
| Device identifier | A random per-install ID generated on your Mac that lets us count activations without identifying the machine. It is not a serial number or a hardware fingerprint. |
Device telemetry
Each licensed device attaches a handful of technical details to its periodic licence check:
| Field | What it is |
|---|---|
| App version | e.g. 1.0.4 |
| First-seen version | The version at the time the device was first activated, frozen thereafter |
| macOS major version | e.g. macOS 14. No minor or build version. |
| CPU architecture | arm64 or x64 |
| Language | Primary language code, e.g. en or de. Not your region. |
| Release channel | stable or beta |
| Engine pack version | The version of the on-device ML runtime pack you have installed |
| Capabilities snapshot | Whether the browser extension is paired, whether the shell helper is installed, which macOS permissions you’ve granted, and which integration types you’ve configured (types only, for example slack or webhook, never recipients, endpoints, or message content) |
We use it to decide when to drop support for older macOS versions, see which builds to prioritise, and figure out issues that only affect certain setups.
Retention: 180 days from the last time that device contacted us. After that we wipe the telemetry fields. The device activation record itself stays until the licence is deleted, so your device-limit count stays accurate.
Legal basis: legitimate interest (GDPR Art. 6(1)(f)). It’s used only to keep the service running and to improve it.
App help lookups
When you watch an app or a public website, Oversee looks for that app’s public help pages and manuals, so it can read an unfamiliar message on screen correctly. Your Mac does the looking and keeps what it finds for up to 30 days. To make this faster for you and for other people, our server also keeps links to those public pages:
| What we receive | What it is |
|---|---|
| A one-way fingerprint of the site | A salted hash of the main domain of the page you watch (for example the fingerprint of stripe.com). Never the name, a sub-domain or the page address. For a native app, the same kind of fingerprint of its bundle ID or name. |
| Links to public documentation | The address of the app’s public help site, docs file or source-code archive, which your Mac confirmed it could open without signing in. We keep the links only, never the pages. |
| Which account confirmed a link | A keyed hash of your account, used only so that two different people have to confirm a link before we hand it out. It can’t be turned back into your account without a secret that stays on our server. |
Nothing you see on screen, no page content, no sub-domain or page address, and no watch rule or name is sent. Before anything is looked up, your Mac checks that the site’s front page opens without signing in, and it never looks up banking, health, government, email, messaging or login pages. When our server gives your Mac a link, your Mac opens it itself and checks it again before using it. Because the request comes from your signed-in account, our server can see that your account asked about that fingerprint at that time.
Retention: a link is handed out until 90 days after it was last confirmed, and stops sooner if three different people report that it no longer works. The record of which account confirmed a link is deleted when you delete your account.
Legal basis: legitimate interest (GDPR Art. 6(1)(f)). It makes Oversee read unfamiliar app messages correctly, and it reaches us only as a fingerprint of a public site’s domain.
The iOS Companion
The optional iOS Companion app shows you a notification on your iPhone when a watch on your Mac fires. To deliver that, we keep a small push-device record per phone:
| Data | Purpose |
|---|---|
| Device public key | Generated on your iPhone the first time you pair the companion, and stored in the iOS keychain, device-only, not synced to iCloud. The matching private key never leaves your phone. |
| APNs device token | Apple’s token for your phone, used to address the push through Apple’s push service (APNs). |
| Platform | Recorded as ios, so we know which client format to send. |
Each notification is encrypted to your phone’s public key (HPKE, the same construction Apple’s own CryptoKit uses) before it leaves our server. Apple’s push service only ever handles the encrypted envelope. It cannot read the notification’s title or body. Your iPhone decrypts it locally, with the private key that was never sent anywhere.
The encrypted payload is the watch’s notification title and body, so whatever your watch was configured to summarise (OCR text, an AI result, etc.) can end up in it, the same as any other Oversee notification.
On your iPhone, delivered notifications are kept in a local alert history, capped at the 200 most recent, stored in an app-group container shared with the notification extension and the iPhone widgets. Not synced to iCloud, not sent to us. The companion has its own setting to hide notification detail on the lock screen.
Telling you when your Mac stops watching. While a watch is set to alert your phone, your Mac checks in with our server every couple of minutes. We keep one record per Mac: its install ID, whether it is armed, when it last checked in and when we last alerted you, and, if your Mac said so as it went to sleep or shut down, that reason. If a Mac stops checking in, that is how we know to send you a “your Mac stopped watching” notification. The text of that alert, and the list of your watches (names and states) that the app shows on its Home screen, are encrypted to your phone and replaced at every check-in. We hold only the latest ciphertext, and we can’t read it. The record is deleted when your last alerting watch ends or Oversee quits cleanly, 7 days after we alerted you, and with your account. When you tap Stop or “I’m on it” on an alert, that instruction waits for your Mac with its install ID, the watch’s opaque ID, the action and the time. It carries no watch name and is deleted once your Mac picks it up, or after 24 hours.
Looking at a watch from your iPhone. Once you scan a code on your Mac with the companion, the two are paired and you can ask your Mac for a picture of what a watch is looking at: from an alert, from a watch’s card in the app, as a small preview that refreshes about every minute while the app is open, or as the last picture a finished watch saw. Your Mac takes the picture at that moment (or uses a frame its own history already holds), crops it to the region you watch, and encrypts it so only your paired iPhone can open it. It passes through our server in memory, for at most a minute until your phone collects it. We can’t read it, we never write it to disk, and we never keep it. We can see that a request happened and roughly how large the answer was, not what it showed. Every request is signed by a key that stays in your iPhone’s Secure Enclave, so only that phone can ask, and the app asks for Face ID or your passcode once per visit. On your iPhone the picture stays in memory only, is cleared when you leave the app and is never saved to your photos or a backup. Your Mac shows a notice each time and keeps a log of the most recent 100 requests in Settings. You can turn this off for any watch on your Mac, remove a paired iPhone at any time, and a phone left unused for 60 days is unpaired automatically.
Widget previews (off by default). The companion has a “Pinned watch” widget. If you turn on its Show preview option, that widget can ask your Mac for a small picture of one watch about every 15 minutes, without Face ID, so anyone using your unlocked iPhone can see it. The picture is encrypted the same way. Your iPhone keeps only the latest small copy for each widget, for at most an hour, in storage excluded from backups, and deletes it earlier if you turn the preview off, remove the widget, unpair or sign out. Pictures never appear on the lock screen, and are not drawn at all while the companion’s Face ID lock or “Hide alert details” is on. Your Mac tells you when a widget is showing a watch and lists these widgets in Settings, where you can remove them one by one or all at once.
Pictures on finished watches (off by default). If you turn on Picture on finished watches in the companion’s Settings, your Mac attaches a small picture of what the watch last saw to the notification that tells you it finished. The picture is taken on your Mac and encrypted so only your paired iPhone can open it. It passes through our server’s memory until your phone collects it, or for 10 minutes at most. We can’t read it and never store it. Your iPhone shows it with the notification, and iOS keeps it there until you clear that notification (and it can appear on an Apple Watch that mirrors your iPhone’s notifications). The companion never adds it to its alert history. Nothing is attached while the companion’s Face ID lock or “Hide alert details” is on, or for a watch with peek turned off. Your Mac lists each iPhone that has this on in Settings, where you can turn it off. Email alerts never include a picture.
Legal basis: contract (Art. 6(1)(b)). We can’t deliver a push notification, tell you a Mac has stopped watching or pass you a picture without these.
Retention: the device record is removed when you unregister the companion, when Apple tells us the push token is no longer valid (for example after the app is deleted), or when you delete your account (it’s included in the account-deletion cascade). We don’t run an automatic expiry on a phone that stays installed but unused, so it stays on file until you unregister it or delete your account. Retention for the status record and in-transit pictures is in the table under How Long We Keep Things.
Onboarding and support requests
When you submit the in-app support form during onboarding or reach us later, we receive:
- Your email
- The message you typed
- The app version
- An optional diagnostics object (e.g. which permissions are granted, whether sign-in has succeeded on your machine) only if you choose to include it in the onboarding support form
These go straight to our Crisp support inbox so we can reply. We do not keep a separate copy in our own database.
Language waitlist
If your language isn’t supported yet and you sign up to be notified, we collect:
- Your email (stored encrypted, alongside a one-way hash used only to avoid duplicate sign-ups)
- The language code you’re interested in (e.g.
de,fr)
Legal basis: consent (GDPR Art. 6(1)(a)). You give us your email specifically so we can tell you when that language ships, and for nothing else. We never use a waitlist address for marketing, and joining the waitlist does not create an account.
Retention: whichever comes first, 30 days after we ship that language, or 2 years from sign-up. After that the record is deleted automatically.
Erasure: we send you a single notification email when we ship that language, and every notification email contains a one-click removal link. You can also ask for removal at any time by emailing [email protected], and we will delete the record whether or not you have an account with us.
Diagnostic and support data
| Data | Purpose |
|---|---|
| Crash and error reports (self-hosted GlitchTip) | Keep Oversee stable and fix bugs. When an error is captured, we send the app version, OS version, a redacted stack trace, and recent breadcrumbs. Breadcrumbs are short diagnostic events from the last few minutes. They include which integration types you used (for example slack or email), how long they ran, and any failure messages, but never recipients, endpoints, or message content. We don’t send watch content, screenshots, OCR text, command lines, file paths, or user-generated text. Our scrubber strips detected emails, phone numbers, and user home paths from error text before it leaves your Mac. The app submits these automatically from both the main process and the renderer, to the same self-hosted GlitchTip at errors.oversee.so. You can turn this off in Settings → Send anonymous error reports. |
| Debug reports (troubleshoot) | When you click “Send debug report” in the Support panel we upload a bundle that includes recent diagnostic logs (text only, no watch frames), a snapshot of your Oversee configuration, the app version, and any notes you add. You choose when to send these. We never auto-upload. |
| Debug reports (flagged watch) | If you flag a result in an active watch (to say “this was wrong”), we upload the watch’s run state, its event log, up to 50 recent keyframes (cropped screenshots of the watched region), and your optional 200-character note. Used to fix false triggers. Those screenshots can contain anything that was visible inside the region you were watching, including information about other people or other apps, and the dialog says so before you send. |
| Debug reports (flagged rule draft) | If the AI rule generator produces something wrong (wrong comparison, wrong value, wrong logic, missing or extra conditions, etc.) and you flag it, we upload: your original goal text (the prompt you typed), the raw model output that produced the rule, the generated rule itself, your reasons and any optional 200-character note, and recent diagnostic logs. No screenshots and no watch frames. The bundle itself contains no account information, but the upload is authenticated, so we know which account it came from for support follow-up (the dialog says so explicitly). Before you click Send, the dialog tells you exactly what is about to be sent and offers an “Export locally” button so you can save and review the bundle first. Bundles are capped at 25 MB. We only use these to improve the rule generator. |
| Support conversations (Crisp) | When you message us through the in-app chat or by email, we store the conversation so we can help you. |
| IP addresses | Briefly logged by our server for rate-limiting and abuse prevention, then discarded. Not kept against your account. |
All three kinds of debug report are uploaded to Bunny CDN storage and deleted 60 days after upload by an automatic sweep. Nothing in a debug report is used for anything other than fixing the problem you reported.
What Stays on Your Device
These never leave your computer unless you explicitly share them. After you uninstall, we can’t delete them for you. You control them directly.
| Data | How long it’s kept on your device |
|---|---|
| Your watches and their configurations | Kept until you delete the watch or uninstall |
| Screen captures, OCR text, frame analysis | Held in memory only while a watch is running. Not saved anywhere. |
| Watch history frames | Compressed frames of the regions your watches had in view, kept so you can review what happened while you were away. The store has a size limit, and once it is reached the oldest watches are cleared first. |
| Classifier training crops | See below |
| iPhone pairing and peek log | The iPhones you paired (their public keys) are kept until you remove them or 60 days pass without use. The log of requests keeps the most recent 100. |
| Shell integration state | When you install the optional shell helper, small per-shell JSON files are written to ~/Library/Application Support/Oversee/shells/ and auto-delete when the shell exits. We never transmit these. |
| Diagnostic logs | Rolling 30-day window. Older logs are deleted automatically. Excluded from debug report uploads unless you explicitly include them. |
| Settings and integration configuration | Kept until you uninstall Oversee or clear application data |
| Temporary screenshots and caches | Cleared automatically when the app quits |
Classifier training crops
When Oversee’s on-device classifier looks at a region, it can save a small cropped screenshot of what it saw so the model can be improved. Alongside each crop it stores the label the classifier chose, its confidence score, the name of the app and the window title the crop came from, and any accessibility text the app published about that element.
These crops never leave your Mac. They are not uploaded, not sent to us, and not used to train anything on our servers. They exist so that a future version of the app can run a better local model on your own machine. The only way one of these images can reach us is if you separately choose to flag a watch, and that upload is described above.
They are pruned automatically: crops older than 180 days are deleted, and the store is capped at 5,000 files, whichever limit is reached first. You can delete them all at any point by clearing Oversee’s application data.
AI-Assisted Rule Drafting
When you create a watch, Oversee first tries to draft its rule with a small model that runs entirely on your Mac. For requests that model can’t confidently handle, and only while cloud drafting is enabled, Oversee can instead ask a cloud AI model — currently Google Gemini — to draft the rule. This happens once, when you create the watch. It never happens again while the watch runs, and it is the only point in Oversee where content from the region you’re watching can leave your Mac before you choose to send it somewhere yourself.
What’s sent
One request, containing:
- The app name and window title of what you’re watching
- The coordinates of the region you selected (position and size)
- The on-screen text of that region (OCR or accessibility text)
- Any structural hints Oversee detected about that region (for example, that it’s a list of records, and the names of its fields)
- The goal you typed
- A screenshot of the window you’re watching, with the region you selected outlined on it
The screenshot is sent by default. Oversee takes it when you select the region and sends it with the draft request. If the first draft doesn’t work, Oversee makes one more request, which includes the same screenshot. It is the whole window, not just the region you selected, so it can contain anything visible in that window, including content outside your region and information about other people. It is sent as captured: Oversee doesn’t blur or redact anything in it. There is currently no setting in the app to turn it off.
Who sends it, and to whom
- If your Mac has its own Gemini API key configured (a developer/advanced setup, not something the app’s Settings currently expose), Oversee calls Google’s Generative Language API directly from your Mac with that key. We never see this request. It has no contact with our server at all.
- Otherwise, if you’re signed in, the same request goes to our server instead, which forwards it to Gemini using our own key and streams the response straight back to you. Our server does not read, log, or store the prompt or the response it relays. This is the path almost everyone using this feature is on.
- If neither applies, the watch drafts locally and no request leaves your Mac.
What we keep
Requests sent through our server (the second case above) leave one trace with us: a record that a call happened — a timestamp, which model, and how long it took — so we can enforce plan limits and diagnose problems. It contains no prompt content and no screenshot.
If a cloud draft made through our server succeeds, we additionally keep the goal text you typed and the rule Oversee generated from it, so we can measure and improve the drafting model over time. We do not keep the on-screen text, the structural hints, or any screenshot — those are used only for that one request and then discarded. If you choose to give feedback on a generated rule (thumbs up/down and an optional note), that feedback is stored linked to the same record.
All of the above is deleted immediately if you delete your account. Currently there’s no automatic time-based deletion on top of that — see the retention table below.
Direct-mode requests (your own API key) never reach our server, so none of this applies to them — Google is the only party that sees that request, under its own terms.
Google’s role
Google processes cloud-drafting requests under its own terms, whether they reach it directly from your Mac or relayed through our server. See Google’s Gemini API terms if you want to know how Google itself handles that content.
Your Role and Ours
For your account, licence, and device telemetry, Serial Labs Ltd is the data controller.
For the content of the things you watch and anything your configured actions send to third parties (see below), you are the data controller and Oversee is a tool you run on your own Mac.
If you’re a business or institutional user and need a Data Processing Addendum (DPA) under Article 28 GDPR, email [email protected].
Integrations and Actions
Oversee can notify you or take action through services you choose. All integration code runs on your Mac, and nothing you send through an integration passes through our servers. The sections below explain each category so you can see exactly who is doing what.
Direct integrations (you bring the endpoint and credentials)
These send data straight from your Mac to services you’ve set up yourself:
- Custom webhooks, Slack, Discord, Telegram, Pushover. You provide the webhook URL or bot token. When a watch fires, we render the message template on your Mac and send the result (and optionally a screenshot) directly to the endpoint you configured.
- OpenAI and LM Studio (AI transforms). You provide the API key, or for LM Studio the local server URL. Your screenshot and prompt go directly to the service you selected.
For these, you are the data controller. The data goes from your Mac to a destination you chose, we never see it, and we have no way to read or recall it later. The service you send to handles it under its own privacy terms and its own retention. We have no contract with it on your behalf, and we cannot delete anything from it for you. You should read the privacy policy of any service you connect before sending sensitive content through it.
Your API keys, bot tokens, and webhook URLs are stored on your Mac, in the macOS keychain or in Oversee’s local configuration. They are never sent to our servers.
Bundled Apple Shortcuts (we provide the shortcut, you provide the recipient)
The Email, iMessage, WhatsApp, Apple Notes, and Apple Reminders integrations use Apple Shortcuts files that ship with Oversee and get installed into your Shortcuts app when you enable the integration. When a watch fires:
- Your Mac renders a message template with the watch details you chose (name, app, window title, duration, OCR text, AI result).
- The Oversee-authored shortcut receives that text (and optionally a screenshot) as input.
- The shortcut dispatches it to the relevant Apple app (Mail.app, Messages, Notes, Reminders) under your own Apple account.
Oversee authored the shortcut, which means we decided what technical fields the shortcut can pick up from each watch. You decide the recipient, the template content, and when it triggers. The data flows on your Mac and to your own Apple apps. It does not reach our servers. You are the data controller for what gets sent and where. Apple handles the resulting message, email, or note under its own terms.
AI analysis prompts
For the AI transform modes, Oversee ships default prompts like “briefly describe what you see” or “what changed between these two images”. You can keep the default or write your own prompt. The prompt and your screenshot go to the engine you selected:
- Apple Intelligence (native or via shortcut). Runs on-device, with Apple’s Private Cloud Compute for harder queries. Governed by Apple’s own privacy policy.
- LM Studio. Runs on
localhoston your Mac. Nothing leaves your machine. - ChatGPT (via Apple Shortcuts). Uses Apple’s built-in ChatGPT shortcut under your own ChatGPT account. Governed by OpenAI’s privacy policy.
- OpenAI (API transform). Uses an API key you provide. Data goes directly from your Mac to OpenAI’s API. Governed by OpenAI’s privacy policy.
We do not see the prompt or the response for any of the above.
Automated Decision-Making
Oversee uses local algorithms and on-device ML to decide when a watch has triggered (for example, “the download finished” or “the text now matches”). These decisions only affect what happens on your Mac. They don’t produce legal or similarly significant effects on you. We do not carry out profiling in the GDPR sense and we do not feed any of this into a broader automated decision system.
Browser Extension
The Oversee Chrome extension is optional. It exists only to let Oversee capture regions of a web page for watches you create. It:
- Runs entirely on your device
- Sends captured browser frames to the Oversee app only over a local WebSocket on
127.0.0.1:17373, authenticated by a random per-install token. That connection never leaves your machine, and there is no network path from the extension to us. - Never connects to our servers, and sends no browser frame, page content, URL, or browsing history to us or to anyone else
- Reads a page only while a watch you created is running against it
If you don’t install the extension, Oversee still works for native-app watches.
Legal Grounds (GDPR)
We process your data based on:
| Purpose | Legal Basis |
|---|---|
| Creating and managing your account | Contract (Art. 6(1)(b)) |
| Licence activation and validation | Contract (Art. 6(1)(b)) |
| Billing and payments | Contract (Art. 6(1)(b)). Handled by Paddle as an independent controller. |
| Device telemetry | Legitimate interest (Art. 6(1)(f)). Service operation and compatibility planning. |
| App help lookups (a fingerprint of a public site’s domain and links to its public help pages) | Legitimate interest (Art. 6(1)(f)). Reading unfamiliar app messages correctly. |
| Push notification delivery (iOS companion) | Contract (Art. 6(1)(b)). We can’t route a push to your phone without a way to address it. |
| Error tracking and crash reports | Legitimate interest (Art. 6(1)(f)). Service stability. |
| Customer support and debug reports | Legitimate interest (Art. 6(1)(f)). Responding to your request and improving the product. |
| Abuse prevention (rate-limiting, etc.) | Legitimate interest (Art. 6(1)(f)) |
| Product update emails (operational: new versions, breaking changes, security advisories) | Legitimate interest (Art. 6(1)(f)). Sent to active users so the product they paid for keeps working. Every email has a one-click unsubscribe; opting out is honoured permanently. |
| Marketing emails (promotional: announcements of new products, offers) | Consent (Art. 6(1)(a)). Captured at Paddle checkout via an explicit opt-in checkbox. Default off. Every email has a one-click unsubscribe. |
| Sign in with Apple / Google | Consent (Art. 6(1)(a)) at the point of provider authorisation |
| Language waitlist | Consent (Art. 6(1)(a)). You submit your email specifically to be notified, and for nothing else. |
| Cloud-assisted rule drafting (proxy mode) | Contract (Art. 6(1)(b)). It’s how we deliver the specific watch you asked Oversee to create, the same basis as push notification delivery above — not a background operational interest. |
Where Your Data Lives (Our Subprocessors)
Core infrastructure (EU-based)
- Scaleway (Paris, France). Server hosting, Postgres database, and outbound transactional email (magic-link codes). Runs our API under
api.oversee.so. - Self-hosted GlitchTip on the same Scaleway server, for error tracking. No third-party processor sees these events.
Website hosting
- Cloudflare. Serves the oversee.so website, including this page, from Cloudflare’s global network. Cloudflare is a US company. The site sets no cookies and loads no third-party scripts (see Cookies and Local Storage below), but like any CDN, Cloudflare’s edge sees the IP address of every visitor in order to serve the request. We don’t add any further logging or analytics on top of that.
Content delivery
- Bunny CDN (Europe). Hosts the Oversee app downloads, update manifests, the on-device engine pack, and the debug report uploads you explicitly submit. Configured with EU edge nodes.
Essential services
- Paddle. Our Merchant of Record. Handles billing, payment methods, tax (VAT / GST) compliance, and invoicing. Paddle acts as an independent controller for billing data when processing your payment.
- Crisp. Customer support chat and email. Conversations are stored in Crisp per their own retention.
- Google (Generative Language API / Gemini). When Oversee’s cloud rule-drafting step runs through our server (see AI-Assisted Rule Drafting above), we relay your request to Google using our own key, and Google is a sub-processor for that traffic. When your own Mac has a Gemini API key configured instead, you’re dealing with Google directly under its own terms, and we’re not a party to it — the same relationship as the OpenAI/LM Studio integrations described under Integrations and Actions.
Identity providers (not sub-processors)
- Apple. Optional Sign in with Apple. We receive your email, a stable Apple user ID, and (on first sign-in only) your name. Apple is not a sub-processor of ours and does not process data on our behalf. You authenticate directly with Apple, under Apple’s own privacy policy, and no DPA applies to that relationship.
- Google. Optional Sign in with Google. Same scopes: email, name, avatar. Google is not a sub-processor of ours, for the same reason.
We only ever receive what the provider returns at the moment you sign in. We have no access to your Apple or Google account, and we cannot read anything else from it.
Other third parties we interact with (not processors)
- Hugging Face. Hosts the on-device ML model files the Oversee app downloads on first run and when a model is updated. Your Mac fetches those files directly. We send no personal data, no account identifier, and nothing about you in that download. Hugging Face is not a processor of ours and no DPA applies.
- GitHub. Hosts the source of the Oversee Chrome extension and mirrors app update artefacts. App and extension updates are plain file downloads from your Mac. No personal data is sent, so again no processor relationship and no DPA.
- User-configured integrations (Slack, Discord, Telegram, Pushover, OpenAI, custom webhooks, etc.). See the Integrations and Actions section above. You are the controller for data you send to these, we are not a party to it, and each of those services decides its own retention.
How Long We Keep Things
We keep personal data only for as long as we need it for the purpose it was collected, after which it is deleted or anonymised, in line with Article 5(1)(e) GDPR.
On our servers
| Data | How long we keep it |
|---|---|
| Account (email, name, auth) | Until you delete your account. Deletion is immediate and permanent, with no soft-delete window. |
| Licence records | 7 years after last activity. UK tax-record obligation is 6 years minimum, plus a 1-year buffer for disputes. |
| Device activations (identity fields) | Deleted together with the licence when your account is deleted |
| Device telemetry fields (on the activation row) | 180 days from the device’s last check-in, then nulled automatically by a nightly job |
| App help links (fingerprint of a public site’s domain, links to its public help pages) | Handed out until 90 days after last confirmed, or sooner if three people report the link no longer works. |
| App help confirmations (keyed hash of your account) | Deleted when you delete your account. |
| Push device registry (iOS companion: public key, push token) | Until you unregister the companion, Apple reports the token as no longer valid, or you delete your account. No automatic expiry runs on a phone that stays installed but unused. |
| Mac status record (iOS companion: install ID, armed or not, last check-in and last alert times, sleep or shutdown reason, plus an alert and watch list encrypted to your phone) | Until your last alerting watch ends or Oversee quits cleanly, or 7 days after we alerted you. Deleted with your account. |
| Stop / “I’m on it” instruction from your phone (install ID, opaque watch ID, action, time; no watch name) | Until your Mac picks it up, or 24 hours. |
| Pictures and pairing messages passing between your Mac and iPhone (encrypted) | Server memory only, at most 60 seconds, then discarded. Never written to disk. |
| Encrypted picture for a finished-watch notification | Server memory only, until your phone collects it or 10 minutes, whichever is first. Never written to disk. |
| Magic-link sign-in tokens | 15 minutes from issue, then auto-expire |
| Language waitlist entries | Whichever is earlier: 30 days after the language ships, or 2 years from sign-up. Swept automatically. |
| Debug reports of every kind (Bunny CDN) | 60 days from upload, then deleted automatically |
| Error events (GlitchTip) | 90 days from event |
| Support conversations (Crisp), including onboarding support forms | Per Crisp’s retention. See their privacy policy. |
| IP addresses | Held in memory for the length of a rate-limiting window, then discarded |
| Cloud rule-draft usage records (proxy mode: timestamp, model, latency — no prompt content) | Until you delete your account. No automatic time-based expiry currently runs. |
| Cloud rule-draft goal text + generated rule (proxy mode, successful drafts only; never screen text or screenshots) | Until you delete your account. No automatic time-based expiry currently runs. |
| Rule-draft feedback (rating and optional note, if you give it) | Same as the draft record it’s tied to. |
On your device
| Data | How long it’s kept |
|---|---|
| Classifier training crops | Older than 180 days, or once a rolling cap of 5,000 files is reached, whichever happens first |
| Diagnostic log files | Rolling 30-day window |
| Watch history frames | Until the store’s size limit is reached, then oldest first |
| Settings and integration configuration | Until you uninstall Oversee or clear application data |
| Temporary screenshots and caches | Cleared when the app quits |
At our processors
| Processor | Role | Retention |
|---|---|---|
| Scaleway | Server and database hosting (EU) | Only as long as we retain it (see above) |
| Bunny CDN | Debug report uploads, app and engine pack downloads | 60 days for debug reports |
| Crisp | Customer support chat | Per Crisp’s defaults |
| Paddle | Payments and tax compliance (Merchant of Record) | Up to 7 years, to meet their own tax-record obligations |
| GlitchTip (self-hosted) | Error tracking | 90 days |
| Google (Generative Language API) | Cloud rule-drafting relay (proxy mode only) | We don’t retain the request or response; see AI-Assisted Rule Drafting for what we do keep |
International Transfers
Most of your data stays in the EU.
Paddle is a UK company that processes payments globally. Because the UK is outside the EU, Paddle relies on Standard Contractual Clauses (SCCs) and additional safeguards to keep personal data transferred from the EU to the UK (or to Paddle’s sub-processors outside the EU/UK) protected under GDPR-equivalent standards.
Cloudflare is a US company and serves the oversee.so website, including this page, from its global network. That transfer is safeguarded the same way Paddle’s is: Standard Contractual Clauses, which Cloudflare includes in its standard customer terms.
Apple Sign In and Google Sign In involve a handshake with Apple and Google’s identity infrastructure respectively. You authenticate directly with them under their own terms. We only receive the tokens they return.
Google (Generative Language API) is a US company. When our server relays a proxy-mode drafting request, that request leaves the EU to reach Google’s infrastructure. That transfer is safeguarded the same way Paddle’s and Cloudflare’s are: Standard Contractual Clauses, under Google’s own data-processing terms for that API.
Your Rights
You’re in control. You can:
- View or export everything we hold about you. In-app under Settings → Account → Download my data, or via our API (
GET /auth/me/data). The export is a single JSON file covering every server-side data source. - Correct inaccurate information. Email us or update it in-app.
- Delete your account via Settings → Account → Delete account, or by email. Deletion is immediate, with no recovery window.
- Withdraw consent at any time. For both marketing and product update emails, use the unsubscribe link in any email. For waitlist entries, use the one-click removal link in every notification email, or email us.
- Object to processing based on legitimate interest (including device telemetry and error reporting) by emailing [email protected]. We’ll honour the request unless there’s an overriding legal ground to continue. You can also switch error reporting off yourself in Settings.
If you’re in the EU or UK, you also have the right to:
- Access, correct, or delete your data
- Restrict or object to processing based on legitimate interest
- Data portability
- Lodge a complaint with your local data protection authority. In the UK, that’s the Information Commissioner’s Office (
ico.org.uk). In other EU countries, see your national authority’s website.
If you’re in California, you have the right to:
- Know what we collect and where it comes from
- Access or delete your data
- Correct inaccurate data
- Say no to data sales (we don’t sell any)
- Not be treated differently for using these rights
Just email [email protected].
Cookies and Local Storage
Oversee is a native macOS app, so there are no tracking cookies.
Our website (oversee.so) sets no cookies at all. It has no analytics, no advertising pixels, no tracking scripts, no embedded third-party widgets, and no fonts loaded from anyone else’s server. Every asset it serves comes from our own domain. There is nothing to consent to, which is why you will not see a cookie banner.
The Paddle checkout uses strictly-necessary cookies to process your payment and prevent fraud. Those are set by Paddle, on Paddle’s domain, and are not used for tracking.
Inside the app, Oversee uses your macOS Application Support directory and the native keychain (via Electron’s safeStorage) to store your sign-in token and settings locally.
Website Analytics
The Oversee website does not use third-party analytics, advertising pixels, or tracking scripts. If this ever changes, we’ll update this policy and say so before it takes effect.
Security and Data Breaches
We apply data protection by design and by default:
- Encryption in transit (TLS for everything) and at rest (Scaleway / Bunny defaults)
- Email addresses and names encrypted at the application layer in our own database
- EU-based infrastructure
- Strict access controls on production data
- No third-party trackers in the app or on the website
- Per-install random tokens for the local browser-extension bridge, so the extension cannot connect without them
- Your sign-in token is stored encrypted by the macOS keychain
- Minimal data collection, local processing wherever possible
If a personal data breach affects you and is likely to result in a risk to your rights and freedoms, we will notify you and the relevant supervisory authority without undue delay, in line with GDPR Articles 33 and 34.
Age Requirements
Oversee is for users 13 and older. If you believe a child has submitted data, email [email protected] and we’ll delete it.
Do Not Track
We don’t track anyone, so there’s nothing to opt out of. Your browser’s DNT header doesn’t change Oversee’s behaviour, because Oversee doesn’t use cross-site tracking.
Business Transfers
If Oversee or Serial Labs Ltd is involved in a merger, acquisition, bankruptcy, or sale of assets, your data may be transferred to the acquiring entity. We’ll notify you via email or in-app before any such transfer occurs and before your personal data becomes subject to a different privacy policy.
Policy Updates
We’ll notify you of material changes via email or in-app. The current version always lives at https://oversee.so/privacy, which is the link the app and the sign-in flow point at.
EU Representative
For all data protection enquiries from within the European Union, Serial Labs Ltd can be contacted at [email protected].
Our account, licence, and error-tracking data are hosted and processed within the European Union. The oversee.so website is served via Cloudflare, and Sign in with Apple, Sign in with Google, and payment processing via Paddle are the other transfers outside the EU, all disclosed above. Serial Labs Ltd acts as its own representative within the European Union under Article 27(3) GDPR.
Contact
For GDPR or CCPA matters, email [email protected] and include “Privacy Request” in the subject. For all other questions, the same address works.
Our Terms of Service are at https://oversee.so/terms.
Serial Labs Ltd
Registered address: 71-75 Shelton Street Covent Garden London England WC2H 9JQ
This Privacy Policy is governed by the laws of England and Wales. By using Oversee, you consent to the practices described above. We don’t appoint a Data Protection Officer, but we’ll always take your concerns seriously and respond.