Privacy Policy
Effective date: November 16, 2026
Kour is an app for reading, writing and keeping everything you work with, for Mac, Windows, Linux, iPhone, iPad and Android, operated by ParkourLabs, Inc., a Delaware corporation (“ParkourLabs”, “we”, “us”). This policy describes what we can and cannot see when you use the Kour apps and getkour.app (together, the “Service”), what we collect, how we use it, and the choices you have.
Questions or requests: privacy@parkourlabs.io.
Your library is end-to-end encrypted
Everything in your library — notes, documents, spreadsheets, handwriting, canvases, files you import, their titles, tags and folders, your annotations, your conversations with Kour’s assistant, your browsing tabs and reading settings — is encrypted on your devices before it is sent to our servers, with keys that only your devices hold. Our servers store and relay that encrypted data so your devices stay in sync; we cannot read it, and neither can anyone who obtains a copy of our servers.
Searching your library, indexing it and extracting its text happen on your devices, and so does recognising text in images and handwriting where your device can do it. Feeds you follow are fetched by your devices directly from the sites that publish them; which feeds you follow is part of your encrypted library. Your mail accounts are read and sent from your devices, which connect to your mail provider directly. Looking up a reference sends its DOI, arXiv number, ISBN or title from your device to the service that holds it (doi.org, arXiv, Open Library or Crossref). Kour’s assistant runs its steps on your devices, but the AI model it uses runs in the cloud: what the model needs is sent through our servers to a model provider, unless you connect your own provider. Some other features send content too, and a page you choose to publish is stored unencrypted. Those are the exceptions to this section; “Pages you publish” and “When content leaves your device” list them.
Your recovery key. When you set up your first device, Kour shows you a recovery key and asks you to save it. Your other signed-in devices can also let a new device in. If you lose your recovery key and every device that holds your keys, your library cannot be recovered — not by you and not by us. We do not hold a copy of your keys, and we cannot reset them for you.
What our servers can see
To run the Service, our servers keep some information unencrypted:
- Your account: your email address and any other addresses you add and confirm; how you sign in (email code, Sign in with Google or Sign in with Apple) and, for Google and Apple, the account identifier they give us; when the account was created. We store no password. If you connect them: your Kour email address for sending things to Kour, your Telegram account’s identifier and display name, and for each AI app you connect (for example through MCP), which app, what it may do, when it was last used and the name you gave the computer it runs on.
- Notifications: for each device, which kinds of notification it wants and whether it is in use now (a computer’s or tablet’s own word, kept for a few minutes); for a phone or tablet you let Kour notify, the push token Apple or Google gave it. A reminder or finished work told by push goes through Apple or Google sealed for that device: neither they nor we can read it, though we and they see that a push was sent and when, and our servers keep when a reminder is due. Turn each kind off in Settings → Notifications.
- Your devices: each device’s public keys, the name you gave it, its platform (for example macOS or Android), when it was added and was last online, and when and why it was signed out. Sessions are stored only as hashes. We also note each week in which your account was used, only to count how many of the people who joined in a given week still use Kour.
- Encrypted items’ outlines: for each item in your library, its identifier, its kind (note, handwriting, spreadsheet, document, canvas, file, or a part of one, such as a worksheet or a page), its size, when it was created and changed, which file belongs to which item, and its encrypted updates: how many there are and, for each, which of your devices sent it and when — but not its title or content. Your public keys, the history of the public keys you replaced, and your keys wrapped for your devices are stored; they are useless without your devices.
- Sharing: when you share an item, who it is shared with and their role; an invitation keeps the invited person’s email address until it is answered or expires. Links you make keep their size, whether they have a password, how often they were opened and when they expire, and, when you limit a link to certain people, a one-way hash of each of their addresses.
- Billing and usage: your plan, your credits, and per-request usage records (which feature, which model, how many tokens or pages, how many credits) — never what the request contained; how much you download from our servers each month. When you pay by card, Stripe sends us notifications about your purchases, and we keep them as Stripe sends them: what was bought, the amount and its date, the name, email and billing address you gave Stripe, and details of the card such as its brand, last four digits, expiry date and country (never the full card number, which only Stripe holds). When you buy through the App Store or Google Play, we keep the notifications Apple and Google send us about the purchase (its transaction and order identifiers, the product and its dates). If you tell us why you cancelled, we keep your answer.
- Shop orders: when you buy from the shop, what you bought, the name, email address, phone number and shipping address you give Stripe’s checkout, and what Stripe tells us about the payment. Buying needs no account, and an order is not linked to one. We send your name, address, phone number, email address and the order to Printful, who makes and ships it, and we email you about the order and its delivery. We keep orders for our tax and accounting records.
- Invite codes: your own invite code; the one code you used, when, and whether you typed it at sign-in or in Settings; when it came in a link, which kind of link (a shared picture, the line under your mail, a copied link) and that share’s identifier. When you share a picture or copy a link that carries your code, that share’s identifier and kind. A visit to an invite link on getkour.app records its code, the share’s identifier and the kind of link. The line under your mail and your share pictures carry your invite link, so the people who read them can see your code.
Pages you publish
When you publish a note or document as a public page, it leaves end-to-end encryption: its title, its text and a picture of it are stored on our servers unencrypted, so that anyone with the page’s address can read it, and search engines may list it. It stays there until you unpublish it or delete your account; after that, a copy can remain in our file storage’s recovery copies for up to 30 days, and its title in database backups for up to 14 days.
When content leaves your device
Some features need a server or a provider. When you use one of them, only what that feature needs is sent, it is processed to give you the result, and we do not keep it. The providers are named on our Subprocessor List.
- Kour’s assistant: your message and the context the assistant uses — for example passages of the document you are working on — are sent through our servers to the model provider. If you connect your own model provider, the assistant’s requests go from your device to that provider instead. So that a long conversation is not sent again with every step, our servers keep an assistant conversation in memory for up to one hour after its last step; it is never written to disk. When you let Kour choose the model (Auto), the start of what you asked is first sent to a model that picks between Kour’s models.
- Smaller AI tasks: naming a conversation or a group; translating; rewriting a passage; outlining a document (the lines that may be its headings); suggesting where a link belongs (a sentence and the names of your Spaces); routing what you said into the empty entry; summarising a recording (its transcript). Each sends only that text to a model provider.
- What’s Up, once you turn it on: what it reads for the day from your calendar, mail and library, through our servers to a model provider.
- AI search on the new tab: what you typed and the titles and short excerpts of matching items, never their identifiers. Web search on the new tab runs only when you choose it; the icons of the pages it finds are fetched by Kour’s servers, so those sites do not see you.
- Grouping the sidebar by topic: when Group suggestions are on, each time you come back Kour sends, through our servers to a model provider, the titles of your loose tabs and documents, their site names (never the full address), how long ago each was used, which one opened which, each page’s own description and main heading, and the names of your existing groups with up to 12 titles from each — never a page’s text. If you turn on Describe pages first (where your device has a model for it, or you have chosen your own model for small jobs), that model writes one line about each page from its text, and the line is sent with the rest; with your own model, the page’s text goes from your device to that model’s provider, not through Kour. Turn Group suggestions off in Settings → Notifications.
- Web search, paper search and reading a web page: your query or the page address.
- Reading PDF pages and scans: unless you choose This device in Settings → AI → Reading scans and PDF pages, the pages whose text Kour reads are sent to a text-recognition provider; a reader you add yourself is reached from your device, not through Kour. Formula recognition sends an image of the region you select.
- Typesetting (Typst and LaTeX compiled in the cloud, where your device has no compiler): the source of that document, and the names of the packages it uses to the Typst package index.
- Dictation: your audio, to a speech provider. To turn what you said into the text you meant, the transcript is then sent to a model provider with what helps it fit where you are typing: the text just before and after the cursor, any selected text, what you copied a moment ago, what you dictated just before, the words you asked Kour to know, and the kind of app you are typing in.
- Recordings and imported audio or video: to transcribe them, the audio is sent to our server, cut at silences in a temporary folder that is deleted when the request ends, and sent to a speech provider.
- Running code in the cloud: the program Kour’s assistant wrote and the files you allow it to read. Each run happens in a fresh isolated machine that is discarded afterwards. A run you allow to reach the internet can send what it reads to the sites it reaches.
- AI apps you connect (for example ChatGPT or Claude, through MCP): an app you approved on your device asks your device; your device answers with what the app asked for and was allowed to see. The request and answer pass through our servers to reach your device; we do not keep them.
- Other agents (Claude Code or Codex, set up in Settings → AI → Other agents): the agent runs on your computer, signed in to your own Anthropic or OpenAI account. What you write in its conversation, the files it works on in the conversation’s folder, and what it reads from your library when you let it (through Kour on the same computer) go from your computer to Anthropic or OpenAI under that account and their terms, never through our servers. The conversation is part of your encrypted library and reaches your other devices like any other. To set an agent up, your computer fetches the list of agents from agentclientprotocol.com, the agent from where its publisher puts it out (GitHub or the npm registry) and, when your computer has none, Node.js from nodejs.org.
- Sending things to Kour by email or Telegram: a message you send to your Kour address or to the Kour bot is seen by our servers when it arrives, reduced to its text and attachments, and encrypted for your account; the original is then deleted (if it cannot be processed, within seven days), and a deleted original can remain in our storage’s recovery copies for one more day. Mail to your Kour address is taken only from your own account’s address and from senders you let in; anyone else’s is deleted without being read, and nothing of it is kept. One of your devices, the first with Kour open in the order you choose, then opens what arrived and has your Kour assistant handle it. If no device picks it up within 30 days, it is deleted; we tell you three days before. The daily report and the reminders you ask Kour for wait the same way until their time; a reminder is encrypted on your device before it reaches us. We keep the senders you let in, which of your devices handle what arrives and in what order, and when your daily report is due. Kour’s replies reach you the same way: our servers see a reply while sending it to Telegram or to your email address, and keep no copy of it.
- Connecting Notion or Zotero to import from them: our server holds Kour’s connection secret for that service and passes your sign-in to your device; it does not keep your tokens, and your device reads the service directly.
- Feedback: what you write, the screenshot if you attach one, and where you were in the app (the page, the app’s version, your device and language, and the identifiers of the requests just before), so the team can answer. Feedback reaches the team by email.
On-device recognition on phones and tablets
Kour uses Google’s ML Kit to recognise handwriting on iPhone, iPad and Android, and on Android also to recognise text in images, identify languages, translate on the device and scan documents. Your images and text are processed on the device. ML Kit itself sends Google diagnostics and usage data about its own work — the device and app, performance, the size of its input and output, errors, and the languages recognised or configured — under a per-installation identifier. Google states that it does not pass this data to third parties. It is the only third-party SDK in the apps that sends data of its own.
What we collect about how the Service works
There are three kinds, kept apart:
Diagnostics. When something breaks, the apps send an error report: what kind of error, a technical stack trace, which part of the app, the app version and the kind of device and operating system, the last 30 steps before it (the screens you moved between, the requests made and the names of the controls you used), and figures such as memory in use and, for a failed assistant step, the model and tool. An error’s own message is kept only where the programming runtime wrote it; a message the app wrote, which could quote a title, a file name or other text of yours, is reduced to the kind of error. Before a report is stored we also remove email addresses, keys and long numbers, text in double quotation marks, the name of a server a failed connection mentions (unless it is Kour’s), every file path on your device, and addresses outside Kour beyond their site name. When an app crashes, the crash is analysed on your device and only its outline is sent: which functions were running in each thread, and which code modules were loaded; the crash file itself never leaves your device. If you share crash reports with app developers in your device’s settings, Apple or Google may also send us the crash logs they collect. Error reports and performance measurements (how long things took, in the apps and on our servers) are not tied to your account: they are sent without it and carry only identifiers made at random for each request, and one for your device that changes every day, so no two days can be joined. Error reports are kept for 90 days; performance traces for 14 days and performance metrics for 30 days. To group repeated errors, an error’s scrubbed message and its counts are kept while the same error is still being reported. You can turn diagnostics off in Settings → Privacy → Share diagnostics.
Product analytics (de-identified). If analytics are on, the apps send which screen you were on and which action you took (for example “opened a note”, “used the search”), whether the window was visible and in focus, timestamps, a random identifier for each time the app is opened with the order of its events, the platform, and, when you pick a row in the entry, its place in the list and the kinds of rows above it. They never contain titles, text, file names, addresses or search terms, and they are not tied to your account: each device sends them under a random identifier that it changes every month, and our servers record only the country the request came from, not the address. In the United States and Singapore analytics are on by default; elsewhere the app asks you once. You can turn them off at any time in Settings → Privacy → Share usage data, and when you do, what those identifiers sent is deleted. Analytics are kept for 13 months. We keep diagnostics and product analytics in this de-identified form, we do not try to tell who sent them, and we require the same of any provider that stores them for us.
Billing and metering: as described under “What our servers can see”.
We use no third-party analytics or advertising SDKs, no advertising identifiers and no tracking pixels. Google’s ML Kit, described above, is used for recognition on the device, not for analytics or advertising.
Our website
getkour.app shows our product pages, blog and legal pages. When you visit them, we record which page you viewed or which button you clicked, the referring website (its domain only) and any campaign parameters (utm_*), sharing code (ref) and offer in the address. A visit carries the random visitor identifier from a first-party cookie the page sets (30 days), and a code computed from your IP address, your browser and the date with a key only our server holds, so codes from different days cannot be joined. We look up on our own servers the approximate location your IP address indicates and keep the country with the visit; cities are only counted per month, with the city’s region and approximate coordinates and without identifiers, and cities with fewer than five visitors are merged into their country. Your IP address is not stored. Visits by search-engine crawlers are kept apart and not counted.
If you join a waitlist on the site, we keep your email address, what you are waiting for, the site’s language and when you joined, to tell you when it is available.
How we use information
- To provide the Service: syncing your encrypted library between your devices, and the cloud features you use.
- To meter usage and administer credits and paid plans.
- To send transactional email: sign-in codes, answers to your feedback, receipts and important service notices.
- To secure the Service, prevent abuse, and fix problems.
- To understand, in aggregate, how the Service is used.
We do not sell or share your personal information (as California law defines those words), and we do not give it to advertisers or data brokers. Our website and apps do not respond differently to a browser’s Do Not Track signal; they do not track you across other companies’ sites or apps in the first place. For the same reason, a Global Privacy Control signal changes nothing: there is no sale or sharing to opt out of. We do not use your content to train AI models, and we ask the model providers we route through not to collect it.
Google Calendar
If you connect a Google account, Kour asks Google for your email address, to read your calendars and events, and to create, change and delete events, answer invitations and invite guests. You see this on Google’s consent screen first, and nothing is read until you agree.
- How it is used. Kour shows your calendars and events and reminds you of them. It changes something when you do so yourself, in the Calendar or by answering an invitation in Mail; when you ask the assistant, which by default goes ahead if your own words asked for that change and asks you otherwise (you can set it to ask every time or not at all, and inviting guests and deleting always ask); and when an organizer moves or cancels a meeting that Kour placed from your mail. Google may email the guests of an event that changes.
- Where it is kept. Your devices talk to Google directly. The sign-in is kept in each device’s secure storage; so that your other devices can use the account, a copy encrypted with keys we do not have is kept with your encrypted library. Events are cached on each device, not on our servers.
- When it leaves your device. If you use Kour’s assistant, or turn on What’s Up, the events they need are sent through our servers to a model provider to give you the result. An assistant conversation stays in our servers’ memory for up to one hour, and a provider may keep what it receives under its own terms; with your own model provider, events go from your device to that provider. What comes of your events, such as a What’s Up page, a conversation, or a recording made during an event (which takes its title and its guests’ names), is part of your encrypted library, and reaches a chat app (Telegram, Slack, Discord) or email only if you have it sent there. If you approve an AI app and allow it to see your calendar, your device’s answer goes to that app, through our servers unless the app is on the same computer.
- What we do not do. We do not sell this data, use it for advertising, or use it to train AI models. What each AI provider has published about keeping or using inputs is on our Subprocessor List.
- Taking it back. Disconnecting the account in Settings → Mail & calendar removes it, its sign-in and its cached events from all your devices, and asks Google to end Kour’s access. If Google cannot be reached then, access continues until you remove Kour at myaccount.google.com/permissions, which you can do at any time. What’s Up pages, conversations and recordings that already used events stay in your library until you delete them, and your library’s encrypted history may keep an old copy of the sign-in.
Kour’s use and transfer of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.
Who can see your content
Nobody at ParkourLabs can see your library: it is encrypted with keys we do not have. We see content only when you send it to us yourself — in feedback, in an email to support — when you publish a page (see “Pages you publish”), and, for the moment a feature needs it, what is listed under “When content leaves your device”.
Providers
The providers that process data for the Service, and what each receives, are listed on our Subprocessor List. AI providers process what you send them under their own terms; the list states what each has published about keeping or using inputs. “We do not keep it” above is about our servers: a provider may keep what it receives for a while under its terms — OpenAI, for one, keeps API requests in its abuse-monitoring logs for up to 30 days, and Anthropic, which runs parts of Kour’s assistant, deletes what it receives within 30 days, or keeps it longer when it is flagged under its usage policy.
Where your data is
Our servers, database and file storage are on Amazon Web Services in the United States (N. Virginia). Diagnostic telemetry is also stored in a ClickHouse database hosted on AWS in the same region. Requests pass through Amazon CloudFront, which has locations worldwide. Email is sent through Amazon Web Services in the same region. Our AI providers are mainly in the United States and Europe; some speech and recognition providers do not publish where they process data. By using the Service you understand that your information is transferred to and processed in these places.
Retention
- Your encrypted library is kept until you delete it or your account. Deleted items are removed from our storage; a file our storage does not delete at once stays queued and is deleted when we try again, every hour until it is gone. Encrypted copies may persist in database backups for up to 14 days and in our file storage’s recovery copies for up to 30 days. When a document is compacted, the version it replaces is kept for 30 days so it can be restored, and its encrypted copy can stay in recovery copies for up to 30 days more. If you start over with new keys, your old keys and the items set aside with them are kept for 30 days, then deleted.
- Billing and usage records: your plan, credits and receipts are deleted with your account. Per-request usage records (feature, model, amounts, credits — never content) and the holds placed for cloud code runs are kept without your account after it is deleted, for our cost accounting. Stripe’s notifications about your purchases are deleted with your account; Apple’s and Google’s, and a Stripe notification that does not name your account (such as a dispute), are kept for 13 months after they are processed. The payment processors keep their own records as the law requires.
- Diagnostics: error reports 90 days, and an error’s scrubbed message and counts while the same error is still reported; performance traces 14 days, performance metrics 30 days.
- Weeks of use: while your account exists; deleted with it.
- Product analytics: 13 months, or until you turn analytics off.
- Invitations: until they are answered or expire; an answered one 30 days more (90 days for one accepted but not yet completed).
- Things sent by email or Telegram that no device picked up: 30 days.
- Feedback (including screenshots): until you delete your account, so we can follow up. The copies emailed to the team stay in its mailbox after that, until we delete them; ask us and we will.
- Website visits: kept for product analytics for 13 months; they contain no name, email address or IP address.
- Waitlist: your email address is kept for up to 12 months.
- Free credits: so that the same person gets Free’s starting credits only once and its monthly top-up once a month, a keyed one-way fingerprint of your email address is kept even after you delete your account: for the starting credits, for as long as Kour gives them; for the monthly top-up, for the 12 months after the month it was used. It cannot be turned back into your address.
- Invite codes: if you used a code, a keyed one-way fingerprint of your email address and sign-in identity is kept for 12 months, even after you delete your account, only so that the same person cannot use a code twice; for the same reason, a code computed from the network address you used it from is kept for 12 months. Who brought whom, and the shares that carried a code, are kept for 12 months as keyed one-way codes made from account identifiers, so that we can count how invitations spread; once an account is deleted, its code names no one. None of these can be turned back into what it was made from. The record that an account used a code stays with the code, so that the code’s limits still hold.
- Server logs: our servers’ own logs are rotated by size and contain no content, email addresses or IP addresses. Records of the network traffic reaching our servers (addresses and ports, no content) are kept for 90 days, and records of our engineers’ administrative sessions for 365 days, for security.
Your rights
You can see, correct, export and delete your data. Your library is on your devices: notes, documents, spreadsheets, handwriting and canvases can be exported from the app in their own formats. You can delete your account in the app (Settings → Account & plan → Delete account); this deletes your encrypted library, your devices, your sessions, your published pages and links, your feedback, and your plan, credits and receipts. What is kept after that is listed under “Retention”, including usage records without your account, Apple’s and Google’s payment notifications for 13 months, invitations you were sent until they expire, the fingerprints described there, and the copies of your feedback in the team’s mailbox; Stripe, Apple and Google keep their own payment records as the law requires them to. For anything else, email privacy@parkourlabs.io from your account’s address and we will answer within 30 days, wherever you live. Because your library is encrypted with your keys, we cannot export it for you or read it to answer a request.
Children
The Service is not directed to children under 13, and you may not use it if you are under 13. The apps do not ask your age. If we learn we have collected personal information from a child under 13, we will delete it.
Changes
We will post any changes to this policy on this page and update the effective date. For material changes we will tell you in the app or by email before they take effect.
Contact
ParkourLabs, Inc. privacy@parkourlabs.io (privacy requests) · support@parkourlabs.io (general support)