Privacy

What we collect

• Your account: your email, your name, a one-way hash of your password if you set one, and, if you sign in with Google or GitHub, the id they give your account. If you turn on two-step verification, your passkeys' public keys and names, your authenticator app's secret, encrypted, and one-way hashes of your recovery codes. The version of the terms you agreed to and when, and a short description of the browser or app each session is on.
• Your connections: the service's address, the connection's label, the account name and id the service reports when you connect (or a name you typed), and its key or sign-in tokens, which are encrypted.
• Your AI apps: the name of each, what it may reach, when it was last used and the addresses it signs in through. The access Kin gives it is stored as hashes.
• The tool descriptions your connected services publish.
• Your permissions: what you set to Allow, Ask me or Block.
• Activity: for each call to run an action, which AI app, which connection and action, the outcome, the time and how long it took. Not the arguments, not the results, and not what was searched for. For each change, and each request that waits for your approval, a one-way hash of its arguments, so it does not run twice.
• The arguments of an action set to Ask me, encrypted, while it waits for your decision.
• The full version of a trimmed result, encrypted, for a short time.
• The result of an approved request, encrypted, for up to one hour after it runs.
• The inputs and step results of a workflow you published, encrypted, for the 15 minutes a run lasts, and notes your AI apps keep in a scratchpad you turned on, encrypted, until they are cleared or expire.
• If you link a phone: its name, app version and type of screen lock as it reports them, the public key it signs your decisions with, and its notification tokens, encrypted.
• If a coding agent on your paired computer asks you for permission through Kin: which agent and tool asked, a summary of the tool's arguments with keys and tokens redacted and never file contents, the name of the repository's folder and its branch, and the agent's session id. Never the folder's full path. Only the person who paired the computer sees it.
• If your AI apps share tasks through Kin: each task's objective, notes and messages, encrypted and sealed to your workspace, which only you and your own AI apps can read. Workspace owners, admins and auditors see that a task exists, never its content.
• If your AI app sends Kin a bug report: what it wrote about the problem, with keys, tokens, email addresses and long numbers removed, encrypted. Your workspace's owners, admins and auditors can read it, and Kin staff read it to fix the problem.
• If you pay: Stripe's customer and subscription ids, the price you chose, the subscription status, when the current period ends, whether it is set to end instead of renewing and when, when a payment started failing, when we emailed you that your free trial was about to end, and when we ended a subscription because its card was issued outside the United States. Never card numbers.
• Request logs: the method, path, status, timing and network address of each request, and for each request from an AI app, its name, version and user agent, the account, workspace and AI app ids, the AI app's name in Kin, the profile, and the action called with the names of its arguments, never their values. Google Cloud's own record of each request also has the full address, your browser's user agent and the page that linked to it.

The arguments and results of actions pass through our server on the way between your AI app and the service. They are not stored, apart from the encrypted cases above.

A public custom server you connect is listed in Explore, by its host name and public tool descriptions and without naming you or including account data, only if you or an AI app you connect shares it. Keep this private starts ticked when you add one, and Never list your connections in the directory, in settings, keeps all of yours out.

Google user data

When you connect a Google service (Gmail, Gmail (send only), Google Calendar, Google Drive, Google Drive (files you pick), Google Docs, Google Sheets, Google Contacts or Google Tasks), Kin asks Google only for the access that service's actions need, and only when you add it.

What we access. Gmail: your messages, drafts and labels. Gmail (send only): sending the messages you or your AI app write, with no access to read your mail. Google Calendar: your calendars, events, and free and busy times. Google Drive: your files' names, details, sharing and text, read only, and the folders and shared files Kin creates. Google Drive (files you pick): only the files you pick in Google's file picker and the files your AI apps create through it, with their names, details and text, and the folders and files Kin creates; never the rest of your Drive. Google Docs and Google Sheets: Google's permission covers viewing and editing your documents and spreadsheets; Kin opens only the ones you or your AI app name. Google Contacts: your contacts, read only. Google Tasks: your task lists and tasks. From every Google connection: the account's email address and Google's id for it.

How we use it. Only to run an action that an AI app you connected asks for, when your permissions for that connection allow it, and to return the result to that AI app. Changes ask you first unless you choose otherwise.

Google's file picker. When you choose Add files on a Google Drive (files you pick) connection, your browser loads Google's sign-in and file picker from Google, and Google shows you your Drive in its own window. Your browser gets a short-lived Google token that can reach only the files you pick. It stays in your browser and is never sent to Kin. Google then lets that connection open the files you picked, and Kin lists them on the connection's page. Nothing from Google's file picker loads until you choose Add files.

What we store. The account's email address and Google's id for it, and the access and refresh tokens Google issues, encrypted. Content normally passes through without being retained. These exceptions are kept encrypted: the full version of a trimmed result, with access expiring after ten minutes; the arguments of an action waiting for approval, cleared when you decide or with access expiring after one hour; the result of an approved action, with access expiring one hour after the action ends; the inputs and step results of a workflow you published, for the 15 minutes a run lasts; and notes an AI app keeps in a scratchpad you turned on, until they are cleared or expire. Expired copies are cleared by periodic cleanup. Backup copies are separate, as described under How long we keep it.

Who receives it. Only the AI app you connected that ran the action. What that app does with the result is governed by its own terms and settings. We do not sell Google user data, use it for advertising, or use it to create, train or improve any AI or machine learning model. Kin itself does not run it through an AI model. People at Kin do not read it unless you explicitly agree that we may view specific data for support, it is needed to investigate abuse or keep the service secure, or the law requires it.

How to remove it. Remove the connection in Kin. That deletes its tokens at once, and when no other Kin connection uses the same Google account, Kin also asks Google to revoke its access. You can remove Kin's access yourself at any time at https://myaccount.google.com/connections.

Kin's use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy (https://developers.google.com/terms/api-services-user-data-policy), including the Limited Use requirements. Its use of information received from Google Workspace APIs also adheres to the Google Workspace API User Data and Developer Policy (https://developers.google.com/workspace/workspace-api-user-data-developer-policy).

Why we collect it

To run the service, to bill for it, and to keep it secure.

We do not sell your data or use it to train models. Kin itself does not run your Google user data, or the results of your actions, through an AI model. To rank actions for a search, Kin runs the search text from your AI app through a small text model on our own server. That text is not kept or sent anywhere else.

Who else handles it

• The AI apps you connect, which receive the results of the actions they run.
• Google Cloud, for hosting.
• Stripe, for payments.
• Resend, for account email such as sign-in codes, password reset links, security notices and the reminder before a free trial ends.
• Apple, to deliver notifications to Kin for iPhone and Apple Watch, only if you link a phone. Apple receives the phone's notification token and each notification: fixed text such as "An AI app is waiting for your approval.", how many requests are waiting, and an id Kin uses to open the request. Pause notices say what was paused and from where, and a message from an AI app names only which app sent it. Live Activities, if you turn them on, also carry the AI app's name in Kin, the name of the action, and counts and times. Notifications never carry an action's arguments or results.
• Google Workspace, for our email, when you write to us.
• Cloudflare, for DNS.
• PostHog, in the United States, for product analytics.
• Google and GitHub, if you sign in with one of them. We receive your email address, your name and the id of your account there.
• Intercom, for support chat when you open it. If you are signed in, Intercom also receives your email address and account id, so we know who is writing.
• Endorsely, to credit the partner who referred you, only if you arrived through a partner's link.
• Google, for its sign-in window and file picker when you choose Add files on a Google Drive (files you pick) connection.

How long we keep it

• Activity is kept for 90 days on Free and for a year on Pro and Team, as What Kin holds describes. A short record of each request that waited for your approval, without its arguments, stays until the workspace is deleted, so a request you answered cannot run twice.
• The arguments of an action set to Ask me are deleted when you decline it or once it has run. If nobody answers, they can no longer be used after one hour, and a cleanup then deletes them, within minutes while Kin is in use and within the hour otherwise.
• The full version of a trimmed result can be read for ten minutes, then the same cleanup deletes it.
• The result of an approved request can be read for one hour after the action ends, then the same cleanup deletes it.
• A coding agent's permission prompt: its summary until the prompt is answered or lapses, at most two minutes, and a record of it without the summary for 30 days.
• A shared task: until seven days after its last change, or three days after it finishes.
• A bug report from an AI app: 90 days.
• Google keeps request logs for 30 days.
• Automated backups keep deleted data for up to seven more days. A manual backup, taken before maintenance, keeps its copy until we delete it.
• Removing a connection deletes its credential.
• Deleting your account removes the account and workspaces you own from our live database. The benefit and audit records described below remain. Backup copies are separate and are not erased immediately by account deletion. Stripe, PostHog and Intercom keep what they already hold, as What Kin holds describes.

Kin keeps creator-benefit and partner commission reconciliation records, and the audit history of complimentary Pro grants, after an account or workspace is deleted. These records can identify creators, customers they referred, and people who received complimentary Pro. They can include account and workspace IDs, partner IDs and codes, a hash of a creator's email address, payment references, amounts and dates, and operator details and notes. These records currently have no automatic deletion period.

Product analytics

We use PostHog, hosted in the United States, to see how people find and use Kin, such as how many visitors sign up and how many go on to connect a tool, and to find and fix errors.

For every visitor, without asking, we send the pages you open (with ids removed, and nothing after a "?" except the campaign tags of a link we shared, such as utm_source=newsletter), the site that referred you, your browser, operating system and device type, and a few milestones: seeing our prices, which part of a page a sign-up button you click is in, starting to sign up and with which plan, signing up and how, a sign-up that did not go through (only a reason from our list, such as "email already used"), connecting a service or failing to (only which one from our directory, or "custom", and a reason from our list), an AI app connecting for the first time and which kind of app it is (only from our list, such as ChatGPT, Claude or Cursor, or "other"), a workspace finishing setup (its first working connection together with a connected AI app), the outcome of the first action an AI app runs, an AI app's action succeeding (at most once a day), the first read and the first change an AI app runs successfully (and whether a change ran after you approved it), an AI app's action succeeding again seven days or more after the first one, deciding a request, creating a profile, reaching a plan limit, choosing to upgrade, starting checkout, subscription changes and deleting an account. When you sign up, the campaign tags and referring site you arrived with are sent with it (through a Google or GitHub sign-in they ride in its short-lived sign-in cookie), so we can tell which links bring people who sign up. These basics use no cookies and nothing stored in your browser.

We also send error reports, from your browser and from our server: the kind of error, where in our code it happened, and its message with addresses, quoted text, email addresses, long numbers and anything that looks like a token removed. Error reports leave out what you typed, the contents of requests, the details of actions and their results.

Milestones and errors are counted per workspace too. A workspace is known to PostHog by an internal number with only its plan, whether a complimentary offer is what makes it Pro, the month it was created, roughly how many connections it has, and whether it is one of our own.

If you have an account, all of this is linked to an internal account number, never to your email address or name. Apart from what you write in feedback or a survey answer, we never send PostHog your email, name, password or tokens, the names or addresses of your connections or profiles, the names of your AI apps (only which kind each is, from our list), what your AI apps search for, the details of any action, or any result. Events from your browser pass through our own server, which keeps only the items above and does not forward your IP address.

Three more things run only if you choose Allow for the current notice and recording scope, wherever you are: session recordings, short surveys, and an id kept in your browser so tests of new designs stay consistent. A survey asks one question at a time, and we receive your answer. Choosing No thanks, or leaving the question unanswered, keeps these three off. Basic page counts, milestones and scrubbed error reports described above do not depend on Allow.

What a recording can show. Movement, clicks, scrolling and timing on allowed pages, and where things sit on them. On our public pages (the home page, pricing, sign in, sign up, asking for a password reset link, and these policy pages), every word and everything you type is hidden before the recording leaves your browser. On signed-in pages, a recording shows only Kin's own fixed interface: the navigation, page and section headings, and a reviewed list of buttons and links, all listed below. Their words come from that fixed list in our code, never from your screen. A recording can also show which of them is selected, unavailable or busy, and whether all AI access is paused. Everything else on a signed-in page is left out and appears only as an empty placeholder with no words and no size: your name, email and workspace, your connections, AI apps and profiles, approval requests with their details and results, activity, counts and dates, anything from Google or another connected service, everything you type, every form field and every dialog. Audio, video, canvases, icons, embedded pages and anything a browser extension adds are blocked too. The number and place of placeholders can still show, for example, how many requests were waiting. On public pages a recording can show the size and position of blocked areas and the length of hidden text.

Signed-in pages and every word they can show. The signed-in pages a recording can cover are Home and the welcome page, Approvals and a single request, Activity, Connections, a connection's own page (only the navigation shows), AI apps, Permissions, Profiles, Add a connection, Computers, Request a tool, Try it, and Usage and settings. A signed-in recording can contain these words and no others: "Approvals", "Waiting", "Others waiting", "History", "Connections to approve", "API connections to review", "Actions to approve", "Nothing is waiting for you.", "No other requests are waiting.", "No past requests yet.", "Approve once", "Always allow this action", "Deny", "Turn on notifications", "Activity", "Refresh", "Show older activity", "Usage and settings", "Plan", "Usage", "Results", "Notifications", "Account security", "Manage security", "Where you’re signed in", "Phones", "Directory", "Workspace", "Delete account", "Privacy", "Save", "Welcome", "Add a connection", "Set what AI may do", "Connect an AI app", "Test it", "Home", "Connections", "AI apps", "Permissions", "Profiles", "Computers", "Build a connection", "Request a tool", "Try it", "New profile", "Keep work and personal apart.", "Pair a computer", "Start over", "Connect", "Add", "Recent activity", "All activity", "Downloads", "Setup", "Review", "Open navigation", "Close navigation", "Feedback", "Chat", "Sign out", "Pause AI access", "Resume AI access", "Pause", "Resume", "Skip to content". Besides them it carries only fixed names from our code that describe layout and state, such as which control is selected, paused or busy, and the address of the page with any id replaced.

What is never recorded. Pages where you manage sign-in methods, passkeys, an authenticator app or recovery codes, or verify it is you; pages where you choose a new password, pair a computer or phone, finish a sign-in handoff, authorize an AI app, review a connection an AI app asked Kin to build or change, or administer Kin; and Downloads. A page without an explicit recording rule is excluded too. The browser stops recording before you leave an allowed page, and our server independently checks each page, rebuilds the recording from the same fixed list and removes everything else before forwarding it to PostHog.

Your versioned choice. This notice is 2026-10-17.1; its recording scope is reviewed-controls-v1. When you answer, we remember only your choice, those two versions, and the time you chose, in this browser's consent receipt. The receipt is signed by our server; it does not include your name, email, account number or the pages you visited. It records an analytics choice, not permission to access a connected account or run an action. Allow lasts for up to 180 days. An expired, earlier or unversioned Allow does not authorize this scope: optional analytics stays off until you choose Allow for the current version. An earlier No thanks stays off.

If you choose No thanks, none of these three runs and we do not ask again automatically. You can change your answer at any time under Privacy choices at the bottom of our public pages, or in settings. Turning them off stops optional analytics in this browser, removes the optional analytics id kept here, and reloads the page. It does not delete recordings or other analytics already received by PostHog. Your choice does not change the choice in another browser. If your browser sends Do Not Track or Global Privacy Control, it sends nothing to PostHog, and we do not ask. Our server only adds one to an hourly count of visits to our public pages, sign-ups started and sign-ups, split by the campaign tag or referring site you arrived with. Those counts hold no id, network address or browser details, stay on our server, and are never sent to PostHog. Milestones our server records for your account, errors on our server, and feedback you choose to send are still sent, without the campaign tags or referring site.

Feedback. Send feedback in the app sends only what you choose to send, when you choose Send: the topic you pick, your message, the page you sent it from, and, only if you allowed session recordings and that page was being recorded, a link to its masked recording. Before your message leaves our server, we remove email addresses, tokens, keys, long numbers, network addresses and anything after a "?" in a web address, and we hide text that looks pasted rather than written, such as quoted results, code, and data in braces or brackets. Our server sends it to PostHog under your internal account number. Since you choose to send it and it does not track you, it works the same whether or not your browser sends Do Not Track or Global Privacy Control. We keep feedback in PostHog to improve Kin. To have yours deleted, write to privacy@kinlab.ai.

Cookies

We set the cookies Kin needs to work: the sign-in session cookie; short-lived cookies that keep a sign-in with an emailed code, Google, GitHub, a passkey or a connected service secure while it is under way, or carry a phone link through sign-in; a sealed cookie, kept for up to 180 days, with one-way hashes of the email addresses this browser has confirmed with a code, so it can still ask for codes while strangers are being limited; and a signed privacy-choice cookie that remembers your optional analytics answer for up to 180 days. That choice cookie holds only your answer, the notice and scope versions, and the time you chose; it is not an analytics id. The app also keeps a few settings in your browser's storage, including your privacy choice. Basic analytics and PostHog set no analytics cookies. If you allow session recordings, surveys and experiments, PostHog's id for you is kept in your browser's storage until you turn them off. There are no advertising trackers, apart from the partner referral cookie described below. If you open the support chat, Intercom sets its own cookies to keep that conversation going, and while the chat is open it sees the address of each page you visit here. Nothing from Intercom loads until you open the chat. If you arrive through a partner's referral link, Kin checks the partner's code with Endorsely, our partner program service, and tells Endorsely about the referral so that partner is credited if you subscribe. Endorsely gets the partner's code from our server, not your address or anything from your browser. Kin remembers that partner in a cookie of its own for 60 days and keeps the partner with your workspace, so it still counts if you subscribe from another device. Nothing from Endorsely loads in your browser, and a browser that sends Do Not Track or Global Privacy Control is never credited to a partner. If you choose Add files on a Google Drive (files you pick) connection, Google's sign-in window and file picker use Google's own cookies, under Google's privacy policy. Nothing from Google loads here until you do.

Your privacy rights

The Free plan is available worldwide and paid plans only in the United States for now, so you may be using Kin from anywhere. Wherever you live, you can use the rights below. Kin Computing, Inc. answers every request.

Write to privacy@kinlab.ai for any of them, and for any other question about your data. We answer within 30 days. So that nobody else can get or change your data, we act on a request only when it comes from, or is confirmed from, the email address on your account.

• Find out what we hold. Your connections, permissions and activity are in the app. Write to us and we will tell you what else we hold about you.
• Get a copy. A download in the app is not built yet. Write to us and we will email you a copy of your account's data, leaving out password hashes, keys and sign-in tokens.
• Correct it. Write to us to correct your name, your email address or anything else we hold about you that is wrong.
• Delete it. In the app, open Usage and settings, choose Delete account, type your email to confirm, and choose Delete my account. That removes your account and workspaces you own from our live database, and ends paid subscriptions for those workspaces immediately. The benefit and audit records under How long we keep it remain. Backup copies are separate and are not erased immediately by account deletion, and other companies keep what they already hold, as that section describes. You can also ask us by email to delete your account. To remove a single connection, open it and choose Remove.
• Say no to optional analytics. Session recordings, surveys and an id kept in your browser run only if you choose Allow. Choose No thanks when we ask, or change your answer at any time under Privacy choices at the bottom of our public pages or under Privacy in Usage and settings. Turning them off stops them in that browser at once and deletes the id kept there. Your answer applies to the browser you give it in.
• Global Privacy Control and Do Not Track. We honor both. A browser that sends either sends nothing to PostHog, is never asked about analytics, and is never credited to a partner. Milestones our server records for your account, errors on our server, and feedback you choose to send are still sent, as Product analytics describes.
• Object to anything else. To object to any other use of your data, write to us. We will stop it where we can, or tell you why we cannot.

We do not sell your personal information, and we do not share it for cross-context behavioral advertising, meaning ads aimed at you based on what you do on other companies' sites and apps. The partner referral cookie described under Cookies is used only to credit the partner whose link you followed, not to show you ads.

Changes to this policy

When this policy changes, the new version will be posted on this page. A change to optional recording scope requires a new Allow before that scope can run. We keep this notice version and earlier ones, notice 2026-10-16.1 and the October 15, 2026 privacy policy and optional analytics prompt, available so you can review what a choice covered.