Draft - needs review by a lawyer. This text is a template. Have it reviewed by a lawyer before publishing, and replace every [bracketed field] with real details. The Turkish version prevails for users in Türkiye.

Privacy policy

Last updated: 20 October 2026 · Version 0.9 (draft)

What we process and why, where we keep it, who we share it with and what we never do. A short summary first, then the details.

In short

We can't access your mail or connected accounts. Gmail, Calendar, Classroom, Drive and Microsoft services connect on your device; their access tokens and content never reach our servers, are not stored there and are not logged.
AI sees only what you allow - and only your own AI. Agenticly has no shared AI of its own. Everything runs on Agenticly Intelligence - our own rules on your data, with nothing sent to any AI - unless you connect your own Claude or ChatGPT (Settings → Your own AI). Then the AI features you allow send text through our server to your own computer, and from there to Anthropic or OpenAI under your own account. Recordings and photos are read on your device and never sent to any AI. See AI.
Notes, settings, your schedule and files you save are stored on our server under your account (some of it encrypted).
Your location is read on your device only when you press “use my location” or, if you already allowed it in your browser, when Today opens to tell which way your commute goes (compared on your device, never sent to us, never tracked in the background); saved places are encrypted; locations and addresses are never logged.
No ads, no selling data, no analytics or advertising cookies. We only count which app features are used (event names, never content) to improve Agenticly.

1. Data controller

Ayhan Kemaloğlu (ALKE), Harman Sk. 39/1, 34153 Bakırköy, Istanbul, Türkiye. Contact: contact@agenticly.live. For the notice under Turkish law see the KVKK notice.

2. Signing in with Google

When you sign in with Google, Google sends us only (OpenID Connect, openid email profile):

your Google account ID (sub - a number that never changes),
your email address and whether it is verified,
your name (and your profile picture's address, if any).

What we don't receive: your Google password, mailbox, calendar, files, contacts or location history. Signing in gives Agenticly access to none of these.

No passwords: Agenticly accounts sign in with Google (or Microsoft where it is offered); we do not ask for, use or store an Agenticly password. If an older account still had one and you sign in with Google using the same verified email address, Google is linked to that account and the stored password hash is deleted.

Confirming it is you: before security changes (such as App Lock settings, sign-in methods, export or deletion) Agenticly asks Google to sign you in again. From that sign-in we only check your Google account ID, that it belongs to your account and when you signed in; nothing else is kept except a security log entry.

App Lock PIN: Agenticly opens with a 4-8 digit PIN you choose. It is never stored in plain text - only a one-way Argon2id hash, which is itself encrypted with our server key (AES-256-GCM, bound to your account), so a copy of our database alone is not enough to try PINs. It is checked on our server one attempt at a time: after 10 wrong PINs in a row or within a day the PIN stops working until you unlock with your password, a passkey or Google, and every wrong attempt appears in your security log. PINs that are easy to guess (like 1234, 0000, 1212 or a year) are not accepted. If you forget it, you sign in with Google again and choose a new one.

3. Connected accounts: Gmail, Calendar, Classroom, Drive, Microsoft

These connections are optional and live on your device:

Permissions are requested only when you connect a service, and only for that service (e.g. read Gmail; read calendar events; your classes and your own coursework in Classroom; in Drive only files Agenticly created or you pick). Sending mail and writing to your calendar are separate permissions and need your confirmation every time (plan blocks, below, follow the blocks you accepted once you turn them on).
Plan blocks in Google Calendar (optional, off by default): if you turn on “Add my accepted plan blocks to Google Calendar” (Settings → Connected apps), Google is asked for one more permission, calendar.app.created: it lets Agenticly make a separate calendar called “Agenticly” and manage the events in it, and gives no access to your other calendars or events. From then on, each time-plan block you accept in the checklist is added there by your browser, directly to Google: the task title (or “Planned block” if you chose to hide titles), the start and end time in your account’s time zone, a hidden key made of the task’s id and the block’s date and time (so the same block is never added twice) and, as the description, only a link back to Agenticly. When you undo an accept, finish the task early or rename it, your browser removes or updates its events; events you add to that calendar yourself are not touched. None of this goes through our server. This browser keeps the switch, the hide-titles choice, the id of the Agenticly calendar and the outcome of the last update (a count and an error code, no titles); it is not synced. Turn it off in the same place; “Turn off and remove the added blocks” deletes the Agenticly calendar with every block in it, and you can also delete that calendar in Google Calendar or remove Agenticly’s access at myaccount.google.com/permissions.
Your browser receives the access token and talks to Google/Microsoft directly. Our servers never receive, store or log these tokens, and your mail, calendar and file contents never reach our servers.
So nobody on our side - including Agenticly staff - can access your Gmail or connected accounts.
Remove a connection in Agenticly under Settings → Connected apps, or directly at Google (myaccount.google.com/permissions) or Microsoft. Removing it deletes the token from your device.

Agenticly's use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.

4. School and work data

School: Classroom courses and your coursework, exam dates, grades (for courses), flashcards and your timetable. What comes from Classroom is read and shown on your device. The timetable, exams, notes and cards you add yourself are stored on our server under your account so they appear on your other devices. When you write or edit a study note, Agenticly Intelligence also makes flashcards from it on your device by our rules (no AI, nothing extra sent anywhere); they are stored with your other cards. For a lesson recording the cards are made from the lesson part of its note. If your own AI structured the recording and study is allowed for it (Settings → Intelligence), your own AI also writes flashcards and quiz questions from that lesson part - once per recording, see AI, the Study row; a switch under Settings → Notes & Voice turns this off and a button on the note makes them on request. They are stored with your other cards and questions too. A card you edit, hide or delete is never changed or made again. Personal notes and notes with “Use with Intelligence” off get none.

Your school and curriculum (optional): whether you go to a state or private school or university, your grade, your school’s name (typed, or picked from the place search - the search text goes to the place search providers below, biased to your city) and where your courses come from are stored with your study settings, under your account. For a state school the national programme (e.g. MEB) is added to your courses automatically. A curriculum document you add (PDF, photo or text) is read on your device by Agenticly Intelligence - the PDF’s text layer, on-device text recognition for photos and scans, and our rules; the file itself is not uploaded or stored. Only the outline found in it (subjects, units, topics, weeks) is stored as your own curriculum and added to your courses; you can edit or remove it in Settings → Home & School. Photos are read only on this device and are never sent to our server or to any AI. Your school’s name is not sent to AI. Reading a photo may be inaccurate; check the result.

From Classroom to School: with Google Classroom connected, homework and exams found in your Classroom notification emails or in your coursework list are added to School on your device and then stored with your account like the ones you add (title, course, due date and time, Classroom link, done state). The ids of the notification emails already handled (ids only, no content) are stored with your account too, so a homework you delete does not come back on another device or after you sign in again. Classroom content is read by your browser; if you turn on the Classroom AI scan and your own AI is connected, the text of Classroom notification emails is also sent to your own AI to read the assignment and its date (see AI).

Work: mail, calendar and files stay in your connected accounts and are processed on your device. When you connect an account that belongs to your employer or school, you are responsible for following their rules; an administrator may have restricted some permissions.

Meetings (optional): you can have Agenticly listen to a meeting and take notes. It never starts by itself: the microphone opens for the meeting only when you tap “Start listening”, and the screen shows that it is recording until you stop. The reading for a voice print and “Test my voice” also open the microphone, each only when you tap its own button, with a level meter on screen while it is open (see Device data, Voice print). Because a meeting has other people in it, their voices and words are processed too: the live text your browser makes while you listen (see Browser transcripts) contains what everyone says, and so does the note that is saved. Tell the people in the room or on the call that you are recording or taking notes - Agenticly cannot do that for you and never claims that anyone agreed; where the law requires their consent, getting it is your responsibility (see the terms). Before every start you tick that you have told them; the time of that tick is saved in the meeting’s note - it records your own statement only, never anyone else’s agreement. If you are under 18, use Meeting - and make a voice print - only with the knowledge of a parent or guardian [lawyer: users under 18; parental consent for biometric data and for recording other people, classmates included; special-category rules]. The recording itself is not kept unless you choose to keep it, and then it stays in this browser only and is never uploaded. What is saved is a note (kind “Meeting”) with the text of the meeting, your own typed notes, how many minutes it was planned, if the voices could be told apart how many seconds each side spoke and, to draw the charts, numbers about the sound and pace of the room for every 30 seconds of the meeting (a very long meeting uses longer steps, up to 10 minutes). These numbers are an energy figure and a clarity figure, words per minute and who was speaking in each short stretch; the first two are worked out on your device from how loud the room was, the background level, the share of the time someone was speaking, how fast people spoke and how sure the browser’s speech recognition was - numbers only, no recording and no text. They cover everyone the microphone hears, other people and a call that comes out of your speakers included, and they describe the room as a whole, not any one person (you can switch the measuring off with “Measure the room’s sound for the charts” on the Start screen). The note is stored in your account like your other notes (so it is on your other devices, is deleted with the note and follows the retention below). Lines are marked “You” and “Others” only if you made a voice print on this device (see Device data, Voice print) and your browser can run the voice check; otherwise they are not marked. The marks are a guess from your voice print and can be wrong. An online meeting in another tab (optional): on a computer with Chrome or Edge you can choose “An online meeting in another tab” on the Start screen. The browser then asks you to pick the tab of the meeting and to share its sound; Agenticly stops the picture at once and keeps only the sound. The sound of that tab (the other participants) and your microphone (you) are mixed on your device into one recording, so the others are heard even through headphones. Lines and seconds are then marked “You” and “Others” by which side the sound came from (your microphone or the tab) - no voice print is used for it, and a mark can still be wrong when sound leaks from your speakers into the microphone. The live text your browser makes still hears the microphone only, so the words of the others are written only if you keep the audio and have the on-device speech model (Settings → On-device AI): after the meeting the whole kept sound is written out on your device, without “You” / “Others” marks on the lines, and the summary is made from that text. Nothing about the shared tab is sent anywhere: its sound goes to the same recorder, is kept only if you chose to keep the audio, and the sharing ends when you stop the meeting or press the browser’s “Stop sharing” button. The summary, decisions, to-dos, open questions and “you could improve” points are written on your device by Agenticly’s rules or - while your own AI is connected and running - by your own AI (see AI, the Meeting row); a meeting never goes to any other AI service. While you listen, a “Say now” card suggests what you could say next by rules on your device; your own AI writes wordings only when you tap “Ask my AI” (see AI, the Meeting row). After the meeting Agenticly can show how it went as charts, strengths, advice and, if lines are marked “You” and “Others”, a note on how you came across: worked out on your device by its rules and, when you tap “Coach me” and your own AI is connected, by your own AI - never by any other AI. This describes how a conversation went and your own part in it. It is not a psychological, medical or HR assessment of you or of anyone: nothing is worked out about another person’s mood, mind, health or character, and the other people are one group (“Others”). The coaching text of your own AI becomes part of the note only when you tap “Add to the note”. [lawyer: whether sound figures of all participants, and communication notes about your own part, are personal data of a special category; whether the rules on emotion recognition at work and in education apply] A presentation or workbook is built on your device when you tap, shown on screen and downloaded; it is never sent to our server or to an AI and is not kept by Agenticly anywhere. It contains what the people in the meeting said, so share it only with people who should have it; your strengths, advice, notes on how you came across and message drafts are left out unless you tick “Include my own analysis”. A message draft is text you copy yourself - Agenticly sends nothing to anyone. The same goes for the follow-up message listing the decisions and who does what by when (also offered on Today the day after a meeting where others took on to-dos): you copy it or open it in your own mail app, and Agenticly never sends it. The owner and day of each to-do are read by rules on your device and you can change them; only the to-dos marked as yours are ticked to begin with. The goal you can type on the Start screen is kept in memory only (not in the note, the draft or the settings) until you close the meeting, and goes to your own AI only when you tap “Ask my AI”. A to-do becomes a task only when you tap “Add selected as tasks”. While the meeting runs, its text and the sound figures counted so far are also kept as an unsaved draft on this device (see Device data).

Lesson check-in: during a lesson on your timetable you can check in (from the lesson prompt, the program list, the School page or the lesson screen). Our server records the check-in time, and the check-out time when you leave early, end the lesson mode or go home; otherwise the lesson is closed at its end. Your first check-in of the day marks you as at school; if a later lesson passes without a check-in, the app asks “Did you go home?” (in the app and, if you turned on notifications, as a notification). If you were in a lesson but could not open the app, you can add it afterwards with “I was there (couldn’t open the app)” - for today’s lessons once they have started and for yesterday’s, only for lessons on your timetable, never for future ones (also for a lesson the app recorded as not attended from a “Did you go home?” answer). It is stored as attended and marked as added later, with the time you said so (our server’s clock) and the minutes of the lesson or the part of it you give; it counts for your attendance and focus rate, never as late, sets “at school” if it was your first lesson and undoes “I went home” if that lesson ran past it. From these records the app calculates, in code and without AI, your attendance per course, minutes present, late check-ins, early leaves and your lesson focus rate. Five minutes before a lesson ends and when it ends the app tells you (in the app and, if you turned on notifications, as a notification, from your timetable only); at the end of the lesson what you typed on the lesson screen is saved to that lesson’s note and a lesson recording is stopped and saved there - on your device and in your notes, as when you save them yourself. Nothing is inferred from your location.

Shared with people you invite (optional, off by default): nothing is shared with another person until you create or enter an invite code on My Space → Sharing. A code works once, for 24 hours; you pass it on yourself (Agenticly sends no invitation), our server keeps only a one-way hash of it, and only a few wrong tries an hour are accepted. Study / work buddy: two accounts linked by a code each see only the other’s display name, whether they ticked “I studied / worked today” (a tick the person makes; it is never worked out from their tasks, focus or anything else), how many days they ticked this week, their weekly goal (1-7 days) and a once-a-day cheer (one fixed emoji). Never task titles, notes, mail, places, times or anything else, and there is no chat. Either side can pause (then neither sees the other’s ticks) or unlink at any time; unlinking deletes the shared record for both. Shared lists: a checklist for up to 6 accounts. Every member sees and can change the items (text, optional due date, done mark), sees the members’ display names and who added or ticked each item (display name only - never an email address). Items are only what members type into the list; one of your tasks is copied into it only when you tap “Share to <list>” for that task. The owner invites and removes members; anyone can leave. When a member leaves or is removed, their access ends and their name is removed from the items; the items they added stay in the list. Weekly summary for a parent or teacher: made on your device from your own week - tasks and homework done, study sessions done with their minutes, flashcards reviewed and exams in the next two weeks; grades only if you tick them; check-in moods never. You see it first and then copy it or open a draft in your own mail app; Agenticly never sends it, does not store it, and no linking is needed. What you share in a buddy link or a list is seen by the people you invited, and what they do with it is up to them. If you are under 18 (by the birth year you gave), a parent or guardian must first confirm by email - Settings → Account sends them one link; until then invite codes cannot be made or entered. [lawyer: users under 18 linking with or sharing a list with adults (a teacher, a parent, a family list) and whether a parent’s consent is needed for that; whether the student-made weekly summary sent to a parent or teacher needs its own notice]

5. What we store

Data	Why	Protection
Account: email, name, Google ID, language, time zone, country/city, preferences	Running your account	Per-account record
App Lock PIN (and, only for an older account that has not signed in with Google yet, its password)	Opening the app; the older password only until Google is linked	Argon2id hash only, encrypted with our server key (AES-256-GCM); the password hash is deleted when Google is linked
Sessions: device type, browser, last seen	So you can see and end them	Server-side; you can end any of them
Notes, tasks (with the steps and planned time blocks you accepted in the checklist and, if you add one, who a task is for - a client, a case or a family member, up to 60 characters; a task you add from a mail keeps that mail’s subject and a link to open it, never the mail’s text), timetable (with your shift rotation, if you add one), recurring lists you make (opening, meals, before the shift: the list names, their items and which items you ticked in the current day, week or shift), the life packs you picked (Settings → Your week), exams, courses (their topics and, if you set them, a target average and an absence limit), cards and quiz questions (including the ones AI wrote from a lesson recording or from the notes and documents you added on the lesson screen), meeting notes (the text of a meeting - which includes what the other participants said - with its summary and to-dos, the numbers about the room’s sound counted for its charts and, if you add them, coaching notes), check-ins (kept only with your explicit consent, asked once before your first check-in and withdrawable in Settings → Intelligence; including the name you give for the person you interacted with most, kept only while a check-in mentions that person), how many cards you rated on each of the last 400 days (a number per day, no card text), your quiz mistakes notebook (for a wrong answer or a right guess: the question and the right answer, the answer you gave - cut short, not kept for a right guess - how sure you said you were, the “what did I mix up?” label and when it comes back; the last 500 are kept) and how sure you said you were on each quiz answer, reminders	Sync and the service	Per account; encrypted at rest with AES-256-GCM, bound to your account and the document. Encrypted: every document stored for your account - notes and calendar entries, tasks, timetable, exams, cards and lessons, your school and curriculum choice and the curriculum outline you added, check-ins, reminders, commute plans, saved places, mail drafts and mail analysis, reports and suggestions, app preferences, uploaded files, notification subscriptions. Not encrypted: the account record (email, name, language, time zone, country/city, profile preferences, your plan and subscription, your birth year if you gave it and, if you are under 18, the email address of the parent or guardian you named with when they confirmed; the App Lock PIN - and any remaining older password - only as an Argon2 hash), the session list and shared caches that belong to no one (news translations, weather, market data).
Program and lesson check-ins: for each block you check in, the outcome (done, partly, moved, not attended), optional details you add; for timetable lessons also the check-in and check-out time (set by our server clock), minutes present, minutes late, whether a lesson was added afterwards with “I was there” (and when you said so), and for the day whether you were at school (time of your first lesson check-in) or said you went home, with your answers to “Did you go home?”	Your program list, the lesson screen, the “Did you go home?” question and notification, and the attendance figures in School and Weekly Intelligence (calculated in code, no AI)	Per account; encrypted with AES-256-GCM; kept 400 days. Never inferred from your location - only what you tap. Undo or change any entry in the program list.
Saved places (home, school, work)	Routes, weather, reminders	Encrypted with AES-256-GCM
Files you upload (blob sync)	Access across devices	Per-account storage, encrypted with AES-256-GCM
Your AI permissions and your choices of what Intelligence and your own AI may use	Consent and your settings	Per account
AI chats (Ask Agenticly / AI Chat): your messages and the answers (and which AI wrote each one), up to 100 chats; a message that no AI could answer is kept as “not delivered”	So you can continue a conversation, also on your other devices	Per account; encrypted with AES-256-GCM. Delete a chat at any time.
AI results kept with your content: note transcripts (and which AI wrote them) and structured sections, PDF analyses, the homework analysis, mail analysis; for a lesson recording, a note of which flashcards and questions were already made from it, so they are not made twice	So the result is not produced again	Per account; encrypted with AES-256-GCM; deleted with the content (mail analyses and reply drafts after 60 days)
Saved links (“Read later”): up to 500 web addresses, each with its title, site, kind, read mark, a link to a note you made from it and the summary you asked for; and, when you open a link for a summary - from this list or by pasting it into the checklist - a copy of the page’s readable text (at most 60 KB per page) with its list items and its own links. From that copy our rules pick out to-dos with dates for you to tick; nothing is added until you do, and no AI is used	Showing your list on all your devices; reading and summarising a page again without fetching it again	Per account; encrypted with AES-256-GCM. A link stays until you delete it (or your account); the copies of page text are deleted after 7 days by the daily clean-up. Not used for anything else.
Voice recordings - only if you turn on uploading them to your account; otherwise they stay on this device	Playing them on your other devices	Per-account storage, encrypted with AES-256-GCM
Homework and exams, also those added from Google Classroom; your school type (state / private, optional) and your own curriculum - including one sorted from your school’s curriculum document (the document itself is read on your device and not sent to our server; only the resulting list of subjects, units and topics is kept); deleted-note markers (id and time); which Classroom emails were already turned into homework or exams (the emails’ ids only)	School, reminders; a deletion reaching your other devices	Per account; encrypted with AES-256-GCM
Feature-use counters: how often a feature was used per day (event names such as “weekly report opened”, never content); your “useful / wrong” feedback on Ask answers	Improving the app	Per account; the counters are kept 120 days. Your “useful / wrong” marks on Ask answers, and the totals of what you use in Ask, stay with your account until you delete it.
Your AI usage list (Settings → Intelligence): task name, time and token count of your recent AI requests, never content	Showing you what your own AI was used for	Server memory only, at most 48 hours
Work style: weekly observed tendencies (computed in code from your activity; not your test answers)	Tailoring the Today order, AI answer length and planning style to you	Per account; encrypted with AES-256-GCM. Reset them anytime with “Reset personalization” in Settings.
Feedback on AI answers: 👍/👎, optional reasons (e.g. “Too long”) and a note of up to 200 characters, the feature, a reference to the answer (a chat message id or a short hash - never the answer text), language, model name when known, time; the last 500 ratings	Your “AI quality” summary in Settings and suggested answer preferences, which apply only after you accept them	Per account; encrypted with AES-256-GCM. Never used to train an AI model. Undo a rating right away, or delete all feedback in Settings → AI preferences.
Answer preferences you add or accept (“What Agenticly knows about how you like answers”; up to 20, 120 characters each)	Sent with your AI requests and added to the instructions for that answer only - so they go to your own AI as part of that request	Per account (all your devices); encrypted with AES-256-GCM. The copy sent with a request is not kept. Edit or delete them anytime in Settings → AI preferences.
Your plan and subscription (only on a server that sells plans): the plan (Free, Formal or Professional), the subscription status (for example active, cancelled, expired, refunded), the renewal and end dates, test or live mode, and Lemon Squeezy’s ids for the subscription, the customer, the first order and the product; and a one-way fingerprint of each of the last 2000 payment notifications Lemon Squeezy sent, so a repeated notification changes nothing. Never card or bank details: Lemon Squeezy takes the payment itself	Giving your account the plan you paid for (the plan is worked out from this on every request, never taken from your browser); showing it, and managing, cancelling or resuming it in Settings → Billing	Part of the account record. When you delete your account, a running subscription is cancelled first and this data goes with the record. Invoices and payment records stay with Lemon Squeezy, which must keep them by law.
Your own AI connection (only if you set it up): which AI you chose (Claude or ChatGPT), when you allowed it and on which version of the permission screen, the paired computer’s name, the pairing time and the CLI versions it reported, and a one-way hash of the bridge token (the token itself stays on your computer)	Sending AI jobs to your bridge and showing its status in Settings	Part of the account record; deleted when you disconnect, when you change or reset your password, when you sign your other devices (or all of them) out, or when you delete your account - the paired computer is cut off at the same moment. Connecting a computer needs a recent confirmation that it is you (the same as for downloading your data), and the connection and its end are entered in your security log (no content). AI jobs are held only in server memory until your bridge answers (at most 90 seconds). A one-time pairing code is also kept only in server memory (and, with its link, in your open Settings page): it works once and expires after 10 minutes; the link carries it after a “#”, which browsers never send to a server. The bridge program itself can be downloaded without signing in: it is open program text with nothing about you in it.
Study / work buddy links (only if you link with someone): the two linked accounts, each side’s display name, the dates the side ticked “I studied / worked today” (the last 14 only), the weekly goal, the date of the last cheer and who paused. Open invite codes as one-way hashes, with who made them, for what and when they expire	Showing each of you the other’s tick, week count, goal and cheer - nothing else	Shared by the two linked accounts; encrypted with AES-256-GCM. Kept until either side unlinks or deletes their account, then deleted at once. An invite code is removed when it is used and at the latest at the first clean-up after its 24 hours.
Shared lists (only if you create or join one): the list name, its members and their display names, who owns it, and the items (text, optional due date, done mark, the display name of who added and who ticked each); for each account, which shared lists and buddy links it is in (ids only)	Letting up to 6 people you invited keep one checklist together	Shared by the members; encrypted with AES-256-GCM. Kept until the owner deletes the list or the last member leaves. Leaving, being removed or deleting your account ends your access and removes your name from the items; the items you added stay with the others.

6. Device data

Device ID: each browser gets a random ID, used to keep per-device preferences (theme, layout, motion) apart and to resolve sync conflicts. It is also kept in a first-party cookie (agl_did, 400 days) that is sent when you sign in, so we can tell a new device and email you a sign-in alert; your account stores a hash of up to 50 such IDs. We don't fingerprint devices.
On your device: browser storage (localStorage) and IndexedDB hold preferences, the offline cache, local files and your connected accounts' access tokens. They stay in that browser; see the cookie policy.
Work-style test (personality test): your answers, the scores computed from them in code, the short/full flag and earlier versions are kept only in this browser (local storage, under a key for your account); they are not sent to or stored on our server. So your other devices won’t have them - you can take the test there too. When an AI request (Ask Agenticly, mail drafts) should follow your style, only derived preferences are sent with that request (e.g. short vs detailed answers, planning style, how you like to learn); they are used for that request only and not stored, and answers and scores are never sent. The weekly observed tendencies computed from your activity are stored on the server, encrypted, and can be reset (see section 5). Not a clinical or psychological assessment; scores are computed by code, not by AI. “Delete profile assessments” in Settings deletes the copy on this device (and any copy an earlier version left on our server). It stays on this device when you sign out, so you do not have to retake it when you sign back in; only deleting your account or this button removes it from this browser. Another account on this device never sees your results. Test answers stored on our server by earlier versions are moved to this device the first time you open the app after the update, then deleted from the server.
Blob sync: with sync on, local files are uploaded to your account so they open on your other devices.
Notifications: if you allow notifications, your browser's push address is stored (encrypted), tied to the sign-in it was made in. It is deleted when you turn notifications off and when you delete your account, and it is no longer used once that sign-in ends (sign-out, “sign out everywhere”, a device you remove, you end its session, you change or reset your password - turn notifications on again on the device you use - or an expired session). Each reminder sent as a notification - its title and short text, for example a homework title, the lesson of an exam, a reminder you wrote or the name of the place to leave for - is encrypted for your browser and passes through the push service of your browser's maker (Google for Chrome and Android, Mozilla for Firefox, Apple for Safari, Microsoft for Edge and Windows), which sees only the encrypted message, your push address and when it was sent.
Calm notifications (optional, all off by default): if you turn them on in Settings → Reminders, our server plans, in code and without AI, up to five kinds of notification from the data stored with us: when to leave (the leave time and travel time from your commute plan), the start of a block you planned for a task (and its first step), “2 minutes to close the day?” at the time you pick, today’s topic in the morning when your exam plan has a session today, and a weekly reset on Sundays at the time you pick. At most 3 are sent a day (reminders you set yourself do not count); none is sent during the quiet hours you choose (22:00-08:00 by default), and those that fall at the same moment are merged into one notification. They go through the push services above as a title and one short line only; unless you turn on “Show task names in notifications” they contain no task name, place name, exam course or topic (for example “A planned block starts now”). Your choices are stored encrypted with your account, together with your reminder settings; your “Snooze 1 h” and “Not today” answers are kept for 3 days like other reminder states. You can turn each kind off at any time.
Meeting drafts and settings: while a meeting runs, its text, what you type and the sound figures counted so far are saved on this device every ten seconds as an unsaved draft (at most three drafts, each dropped after 7 days, removed as soon as the meeting is saved or you discard it), so a crash or a closed tab does not lose it. It is not sent anywhere until you save the meeting. The goal you type on the Start screen is not part of the draft. Your meeting settings (whether you accepted the first notice, keeping the audio, live text, tips, measuring the room’s sound, what is heard - the microphone or an online meeting in another tab - the start offer) are kept in this browser too. If you chose to keep the audio of a meeting, it is stored in this browser’s IndexedDB only - not uploaded, even if uploading recordings is on - and goes when you clear the browser’s data.
Voice print (optional, biometric): if you create one in Settings → On-device AI → Your voice or on a meeting’s Start screen (after you tick an unticked consent box), Agenticly makes a voice print - a short list of numbers that describes how your voice sounds - on this device from about 25 seconds of you reading a text, and keeps only those numbers (the reading itself is never stored) in a database of its own in this browser (IndexedDB). It is made by a built-in method that runs on this device, needs no download and is approximate: it can take other people for you and you for others. It is never sent to our server or to any AI, is not part of an export or a backup and is not synced to your other devices. It is used only in Meeting mode, on this device, to mark which parts of a meeting sound like you: the other people’s voices are compared with it in memory and forgotten, and the meeting’s note keeps only which lines were yours, how many seconds each side spoke and, for the charts, who was speaking in each short stretch. On the Start screen you can also test it with “Test my voice”: for about 5 seconds the microphone listens to the room, other people’s voices included, the sound is compared with your print in memory and forgotten, only “sounds like you / not sure” is shown and nothing is stored. It is a hint, not proof, can be wrong, and is never used to unlock anything or to identify you to anyone. You can delete it with one tap (Settings → On-device AI → Your voice, which is shown also while the On-device AI switch is off, or on the Start screen); it is also deleted when you sign out - including “sign out, keep note files” - and when you delete your account, and a print made by another method or for another voice model revision is not used. The built-in method needs no download and does not need the On-device AI switch to be on; the voice model, where a version includes it, is downloaded like the speech model (see On-device AI models).
Rhythm (your weekly goal, best hours, routine suggestions): to place long work in the part of the day when your finished tasks show work usually gets done, Agenticly reads the times you finished tasks in the last 8 weeks, by rules on this device, and says nothing until it has seen it on at least 5 different days; it describes when things got done, never what kind of person you are. Whether the plan may use it (“Not like this” turns it off), the number of study or work days you aim for each week and which routine suggestions you put off until when are kept only in this browser (localStorage, parked under your account when you sign out); they are not sent to or stored on our server and nothing is sent to an AI for them. The routine suggestion on Today counts your own Program check-ins and ticks of the last 3 weeks; it changes a routine only after you confirm the exact change, and that change is saved like any edit you make in My schedule.
Focus sessions on tasks: when you start focus on a checklist task, the task’s title, its estimate, the minutes you actually spent and how it ended (done, part of it, not this time) are kept only in this browser (the last 200 sessions). They are not sent to our server or to any AI and are not synced. Agenticly Intelligence uses them on this device only to offer a more realistic estimate when you add a similar task with a time (“similar tasks took about 1 h 40 min”) - only with enough sessions, never as a score; you accept it or keep yours. When you tick a task or a step, or add a block to your plan, at the end of a session, that change is saved on the task and syncs like any task. Signing out parks them with your other local data.
Daily briefs (Library → Files): the daily brief presentation (.pptx) is made on your device each morning (or when you first open the app that day, or when you tap “Create today’s brief now”) from data the app already has there - your timetable, tasks, homework and exams, yesterday’s check-in, and, when allowed in Settings → Intelligence, the weather for your city, your commute, your watched markets and the article of the day. It is kept only in this browser (IndexedDB, under a key for your account) for 30 days, then deleted automatically; it is not sent to or stored on our server, and nothing extra is sent to AI for it. It stays on this device when you sign out; deleting your account removes it from this browser. Another account on this device never sees it. Reports, spreadsheets and presentations that earlier versions generated and kept on our server have been deleted; spreadsheets and decks you ask for on the Excel and Slides pages are built for your download and not kept on our server. A meeting’s presentation and workbook are made the same way, on your device when you tap, from the meeting’s note (see Meetings in section 4): they are not kept on our server or on your device and are never sent to an AI.
Close the day, the weekly reset and “My day fell apart” (Today): the first small step you choose in the evening for the next day (shown under Now that morning until you put it away), the up to three priorities you write for a week, with which of your tasks each is linked to, and the steps you accept in “My day fell apart” (the item’s title, its time and minutes, at most six a day) are kept only in this browser (localStorage) for 28 days, then dropped; so is whether you put the weekly reset away for that week. They are not sent to our server or to any AI, and they do not move to your other devices. The changes you make to tasks during these rituals (a new date, “Let it go” to No date, a priority added as a task, a time block you accept) are ordinary task changes and sync like your other tasks (see section 4). Everything here is worked out by fixed rules on your device. “My day fell apart” reads your timetable, routine, open homework, tasks and exams, the plan blocks you already accepted and, only if you connected one, your calendar’s events for today (read on this device, as in section 4); it proposes, you accept, and nothing is written to your calendar or sent by mail.
Empty your head and Good days: what you write in Empty your head stays on this device. The lines you write, how Agenticly Intelligence sorted them (task, worry, idea or note - by rules, on your device) and what you keep (parked worries with the time they come back, ideas, notes) are kept only in this browser (localStorage, under a key for your account): they are not sent to or stored on our server, not synced to your other devices and never sent to an AI. They stay until you let them go or remove them, or clear the browser’s data; when you sign out they are set aside with your other local data for this account. Only a line you accept as a task (“Accept”, “Make it a task”) or save as a note becomes one of your tasks or notes, like anything you add yourself. Good days shows you again the best moment you wrote in an earlier check-in - on Today after your last three check-ins rated the day low (at most every three days, never next to an exam card), on the Good days list, or when you tap “Remind me of a good day”. It is read from your check-ins and nothing is copied; this device only remembers which moment was shown on which day (the last 60).
Working offline: so the app opens and works without a connection, this browser keeps (1) the app’s own files for the current version (code, styles, fonts) in Cache Storage, through our service worker - no personal data, and never an answer from our server about you; an older version’s files are deleted when a new one is stored; (2) the changes you make while offline (tasks, checklist, time plan, notes and the rest), in localStorage, in the order you made them, together with the time of the last sync - they are sent to your account as soon as the connection is back and removed once it accepts them; (3) a copy of your account details as our server last sent them (name, email, language, time zone, settings; not your passkeys), in localStorage, so a reload with no network at all still opens the app. With App Lock or a PIN on, no such copy is kept and the app waits for the connection, because only our server can check your PIN or password. Signing out (after sending what is waiting, or warning you), another account signing in on this browser, or our server answering that you are signed out or locked removes the copy and the waiting changes; the app files stay until a new version replaces them or you clear the browser’s site data.
On-device AI models: if you download them (Settings → On-device AI), the model files are kept in this browser's Cache Storage and the on/off switch in localStorage. They contain no personal data and “Delete from this device” removes them. Recordings and images processed on the device stay on the device (see section 8). If a version includes a voice-recognition model for Meeting mode (see Voice print), it is fetched and stored the same way; the built-in voice method needs no download.

7. Location

Your device location is read only when you click “use my location” and, if your browser already allows location for Agenticly, once each time Today opens: the commute card then checks whether you are near home or near your destination, to show the right leg of your trip. That position is compared with your saved places on your device only; it is not sent to us or to anyone else for this, Agenticly does not store it, and it is never tracked in the background. Agenticly does not ask for the permission there.
Saved places are stored encrypted with AES-256-GCM.
Addresses and coordinates are never written to logs; address searches and coordinates are sent in request bodies, not URLs.
For weather, routes and nearby places, a place's coordinates or your search text are sent - without your name or account - to the provider: Open-Meteo and MET Norway (weather), Google Maps Platform or TomTom (routes, traffic, places), and OpenStreetMap-based services (Nominatim and Photon for address search, Overpass for nearby places, OSRM for routes).

8. AI (your own AI)

Agenticly has no shared AI of its own: the only AI is your own, if you connect one. Without it, nothing is sent to any AI. Agenticly Intelligence (our own rules on your data) does everything it can: Today, the calendar, homework, mail summaries and rule-based reply drafts, note summaries and headings, quizzes and flashcards built by rules, reading timetable text, the checklist (dates, times and lengths in what you type, steps for big tasks, a time plan in your free time, to-dos from a pasted link), summaries of saved web pages, briefs and “why it matters”. Features that need an AI (AI Chat, AI mail drafts and analysis, notes AI, study AI, insights, meeting help, cleaning up a voice transcript, “Search the web and answer”) ask you to connect your own AI and do nothing else. Our server refuses every AI request that is not for your own connected AI. Every result shows its source: “Intelligence” or “AI”.
Your own AI (optional, off by default): in Settings → Your own AI you can let your own Claude (Anthropic) - through the Claude Code CLI signed in with your Claude Pro or Max plan - or ChatGPT (OpenAI) - through the Codex CLI signed in with ChatGPT - do Agenticly’s AI work. A third choice is Qwen (local): Alibaba’s open Qwen3.5 model running on your own computer through Ollama; then no text leaves your computer, and only the model files are downloaded once from ollama.com. Agenticly asks you first and records your yes, its time and which version of that screen you said yes on. You connect with one link or command that the page shows you: it works once and expires after 10 minutes, and the code in the link sits after a “#”, which browsers never send to a server. It works through the Agenticly bridge, a small program you run on your own computer (downloaded as one program from the address Settings shows, on GitHub, which sees your IP address during the download; or run with Node from our site): it pairs with your account, fetches each AI job from our server - the same material the table below lists for that feature, for example the note, voice transcript, email, quiz or chat message the job is about, with the task’s instructions - runs the CLI and sends the answer back. Our server only passes the job through and does not log it. Anthropic or OpenAI then processes it under your own account and its terms, not ours; we never receive your subscription password, login or keys. Your own AI only ever receives text: recordings, photos and scanned pages are never sent to it (they are read on your device, see On-device processing below). Your permissions and the switches under Settings → Intelligence (mail, notes, files, personal, study) still decide what may go to it. While the bridge is not running, features use our rules (AI Chat works differently, see below). A job waits in our server’s memory only until your bridge answers (at most 90 seconds) and is not logged. If you are under 18, a parent or guardian confirms by email before the first pairing. “Measure my AI” in the same screen sends a fixed set of test texts written by us (none of your data) to your own AI and keeps only the score, the model name and the time with your account, never the answers. Disconnect at any time: the bridge stops working at once.
AI Chat on your own AI: the answers come from your own AI, and each stored answer records which AI wrote it. If your own AI cannot answer (not running, busy, too slow or signed out), your message is kept as “not delivered” with a Try again button - nothing else answers it and no reply is made up. In AI Chat your AI may also use Agenticly tools for you: read your timetable, homework, exams, courses and notes (a meeting note too, with what everyone said in it and any coaching notes written into it - like your other notes, unless the note is switched off for Intelligence or AI for notes is off), and add flashcards (a card you already have is skipped) or homework when you ask; it never sends mail or messages.
Voice notes on your own AI: while you record, your browser (or the on-device speech model, if you downloaded it) writes a rough live text draft. When you stop, and only if your own AI is connected and you chose “Your connected AI” under Settings → Notes & Voice → Transcripts (which also turns on the browser transcript described below), that draft text - never the recording - goes to it in parts, with its language and, for a lesson recording, the course and topic names, to be written out as a clean transcript: everything that was said, with recognition errors fixed and punctuation and paragraphs added, nothing summarised or translated. A check on our server refuses an answer that has lost or changed too many of the draft’s words (a summary, a translation or added commentary), and the draft then stays as it is. The clean transcript replaces the draft in your note; the draft stays in the note’s version history on this device only (this browser, the last 30 versions; it is not synced) and the note records which AI wrote it. If there is no draft at all - your browser has no speech recognition and you have no on-device model - there is nothing to send and no transcript is made. If your own AI is not running, the draft stays and the job is tried again later; older notes are not redone on their own, a button on a recording writes the transcript on request - that button works whatever you chose; with “Browser transcript” or “Off”, nothing is sent to your own AI on its own. (In Chrome and Edge your browser’s own speech service may process the audio to write the draft - see Browser transcripts below.) Turning a transcript into a structured note is a separate step: it keeps only the parts about the lesson and lists what it left out by label. For a voice recording of a lesson it may also send the course and topic names and the names of that course’s topics and of the topics of a coming exam that you entered in School, only to help tell what belongs to the lesson (never teacher, room or times).
Lessons: before a lesson is recorded in Study you confirm that your teacher or school allows it; the time of that yes is saved with the recording (your own statement, never anyone else’s agreement). Once a lesson recording has been structured by your own AI and study is allowed for it (Settings → Intelligence), your own AI also writes flashcards and quiz questions from the lesson part of the note only - never the raw transcript or the talk that was left out - and the course name, in one request, once per recording. Each card and question must rest on a quote that is really in that lesson part. They are stored with your other cards and questions; a card you edit, hide or delete is never changed or made again. You can switch this off under Settings → Notes & Voice. On the lesson screen you can also add photos (of the board or a book) and PDFs of the lesson: pages are read on your device (printed-text recognition, or a PDF’s text layer - handwriting is usually not recognised), you check and edit the text, and it is saved as a note of its own. Only when you press “Make cards and questions” does the text of your lesson note and of those documents - never the photos or files - go to your own AI, once per note, with the same rules (each card and question must rest on a quote that is really in the text; private notes are never sent).
Every feature that uses personal content (mail, notes, files, check-ins) needs its own permission (Settings → AI permissions). Without it, that feature works without AI, using rules. If you connect your own AI, your yes to it also counts as the permission for notes analysis, quiz analysis and cleaning a voice note’s draft that you were never asked about - the screen where you say yes lists exactly what is sent. A permission you switched off stays off, and you can change any of them in Settings → AI permissions. Only a yes given on the current version of that screen counts this way: if you connected your own AI before it said so, those permissions are not covered until you connect again and say yes there. Whether a voice note’s draft is actually sent is a separate choice under Settings → Notes & Voice → Transcripts (“Your connected AI”); making that choice also switches the “Voice transcription” permission on, if it was off. “Search the web and answer” always needs its own yes (“Web search”); connecting your own AI does not give it.
Ask Agenticly answers from your data with our rules and sends nothing to any AI; for a question the rules cannot answer, “Ask in AI Chat” opens AI Chat with it. AI Chat (with your own AI) answers your question with the data it needs from your account (for example your timetable, homework or exams, weather, markets, or check-in averages). Each source can be switched off in Settings → Intelligence → “What Intelligence may use”; a switched-off source is never added to the question.
What is sent passes transiently through our server to your bridge and is not logged. The text itself is not kept on our server for AI; the result may be kept with your content (for example a note's transcript or a mail analysis - see What we store). AI chats are stored so you can continue them. So that the same work is not sent to your AI twice, the answer to an identical request is kept in our server’s memory for your account only - for up to 24 hours for most tasks (up to 7 days for reading a Classroom notification email, up to 1 hour for web answers) - and is never written to our database.

What each AI feature sends, and when - only ever to your own AI, and only if you have connected one (see Your own AI above):

Feature	What is sent	When	Permission
AI Chat (Ask Agenticly hands questions to it)	Your question, the recent messages of that chat, and the account data the answer needs: your timetable, homework, check-ins (the ratings and, when you ask about how you felt, your own words in them), the morning brief and the mail highlights in it, and your country, city and time zone by name (never coordinates) - but only the sources allowed under “What Intelligence may use”: a source you switched off (including Location) is never added, whatever you ask. Always, what you told us in onboarding (how your days are structured, what you want help with, your goals). Ask Agenticly itself answers by our rules and sends nothing. It goes only to your own AI, marked as data so that it is not taken as instructions. Without your own AI connected nothing is sent: your message is kept as “not delivered” and no reply is made up	When you send a message in AI Chat with your own AI connected	Your own AI connected; each source can be switched off
Search the web and answer	Only the question you type and any https links you paste. Your own AI’s CLI searches the web from your computer (the search service it uses sees the query) and answers with the pages it cites. Our server then downloads each cited page (at most 8, https only, with limits on size and time, never addresses inside private networks) to check that the quoted text is really on it; a source that cannot be checked is dropped, and facts from the pages are labelled apart from the assistant’s own summary. Nothing from the pages and not your question is stored or logged; the answer is kept in our server’s memory for your account for up to 1 hour so the same question is not sent twice. The sites see our server, not you	Only when you ask on the “Search the web and answer” page, with your own AI connected	“Web search” permission - always its own explicit yes; connecting your own AI does not give it
Mail (analyse, sort, reply)	The email you open or choose. By default Agenticly Intelligence reads mail with our own rules on your device (summary, action, dates, reply drafts) and nothing goes to any AI; only when you press “Ask AI” (or its rewrite options), with your own AI connected, is that one email sent to it. A draft is for you to check and edit: Agenticly never sends a mail by itself	When you ask (“Ask AI”) - never in the background	Your own AI connected; AI for mail on (Settings → Intelligence)
Google Classroom	The text of Google Classroom notification emails (never other emails); your homework titles, courses and due dates for the priority analysis	Only with your own AI connected and AI for study on (Settings → Intelligence), then automatically after you turn on the Classroom AI scan. Otherwise our rules read Classroom mails and order your homework on your device	Classroom AI scan (asked once on this device; can be turned off)
Notes	Note text; the text of PDFs (their text layer, or text recognised on your device from scanned pages) to analyse them; the text of photos of notes and curriculum documents you import, once your device has read it. Never the recordings, photos or PDF files themselves. When a voice recording of a lesson is turned into a structured note it can also send the course and, if you set it, the topic - and, only to help tell what belongs to the lesson, the names of that course’s topics and of the topics of a coming exam in it that you entered in School (never teacher, room, times or other courses)	Only with your own AI connected and AI for notes on (Settings → Intelligence): when you ask, and for new transcripts and PDFs once allowed. Otherwise structured notes, summaries, headings and questions are made by our rules; recordings and photos are read on your device (on-device models) or not at all	“Voice transcription” and “Notes analysis” permissions
Voice note transcript (written by your own AI)	The live text draft of a recording - the rough text your browser’s speech recognition or the on-device model writes while you record - in parts, its language (if known) and, for a lesson recording, the course and topic names (so subject words are spelled right). Never the recording itself. A saved meeting whose lines are not marked “You” and “Others” is handled the same way: its live text is tidied by your own AI only under this same choice (the meeting’s summary is a separate step, see the Meeting row)	Automatically after you stop a recording, only if you chose “Your connected AI” under Transcripts (Settings → Notes & Voice) and your own AI is connected and running - never to any other AI service. If it is not running, the draft stays as it is and the job is tried again later. Older notes are not redone on their own; a button on a recording writes the transcript on request	“Voice transcription” permission - choosing “Your connected AI” turns it on; it also counts as given while your own AI is connected (on the current permission screen), unless you turned it off
Meeting advice, what to say, summary and coaching (written by your own AI)	The text of the meeting, as your browser’s speech recognition or the on-device model wrote it: lines marked “You:” (you) and “Others:” (everyone else together - the marking is a guess and can be wrong), so it includes what other people in the meeting say. Also the meeting title and planned length if you set them, the minutes passed, how many seconds each side has spoken (a count, not a recording) and, for live tips, an agenda if you type one. For live tips only the last few minutes (at most 3,000 characters) go each time. What to say (only when you tap “Ask my AI”): the last 90 seconds of the lines (at most 1,500 characters), the language, the minutes passed and planned, the goal you typed (at most 300 characters), the question you typed under “They just asked” (at most 200 characters) and which kind of wording the app’s rules had offered. Coaching (only when you tap “Coach me”): this meeting’s lines in one request of at most 15,000 characters - your own “You:” lines in full, the others’ lines each cut to 200 characters, the newest part when the meeting is longer - with the language, the minutes, the seconds each side spoke and the figures counted on your device from the sound of the room and the words (energy, clarity, pace, filler words, number of questions). Never the audio	Live tips: automatically while a meeting runs and live tips are on, at most about 30 times an hour, only while your own AI is connected and running. Summary: automatically when the meeting ends, the whole text in parts of at most about 15,000 characters, to write the summary, decisions, to-dos, open questions and points you could improve (the button on the note asks again). What to say: only when you tap “Ask my AI”, at most 20 times an hour. Coaching: only when you tap “Coach me”, at most 12 times an hour, to write strengths, advice, notes on how your own part came across and a message draft; the answer is not saved in the note unless you tap “Add to the note”. Only to your own AI - never to any other AI service. If it is not running, nothing is sent and the app uses its own rules for the summary. Our server keeps neither the text nor the answer. Turn live tips off with the live-tips switch (on the meeting screen and in Settings → Modes), or switch AI for notes off under Settings → Intelligence	“Notes analysis” permission - counts as given while your own AI is connected (on the current permission screen), unless you turned it off [lawyer: whether this yes is enough for analysing a meeting that holds other people’s words, including coaching and what-to-say help]
Study (lessons, quizzes, flashcards, tutor)	The topic and the notes or answers you use, and a short list of questions already made so it does not repeat them. For a lesson recording: only the lesson part of its structured note (never the raw transcript or the talk that was left out) and the course name. For the notes and documents you add on the lesson screen: their text and the course name (photos and PDF pages are read on your device first; the pictures and files themselves are never sent)	When you ask - only with your own AI connected and AI for study on (Settings → Intelligence). Also when you press “Make cards and questions” on the lesson screen (once per note, only for the notes and documents you ticked there). Also automatically, once per recording, after AI has structured a lesson recording: it writes flashcards and quiz questions from the lesson part, each resting on a quote that is really in it (turn it off with the switch under Settings → Notes & Voice, or by switching study AI off). Otherwise questions, cards, reviews and your learning profile come from your notes and answers by our rules	“Quiz analysis” permission for analysing your answers
Files and saved links	A document’s text, to sort, summarise or explain it - only to your own AI. A saved web page is never sent to any AI: our server downloads it for you (https only, never addresses inside private networks) and summarises it by our rules; the site sees our server, not you, and the readable text is kept for 7 days (see What we store)	Documents: when you ask, with your own AI connected and AI for files on (Settings → Intelligence); otherwise our rules summarise the text. Saved pages: when you ask for a summary	AI switched on for files
Morning brief, weekly and other reports	The day's facts: timetable, tasks, reminders, the titles of your calendar entries, check-in scores and, from yesterday’s check-in, your own words (best moment, what could have gone better, your day summary, your note), the sender name and subject of up to six important mails (only with AI for mail on), the patterns and earlier reports the app remembered, weather. Each source can be switched off under “What Intelligence may use” (schedule, school, mail, check-ins, news, location): what is switched off is left out of the brief; the weekly report leaves out your check-ins when they are switched off and otherwise holds only numbers our code counted from your tasks, routines and schedule (counts, percentages, hours)	Only with your own AI connected and AI for personal features on (Settings → Intelligence), when the brief or a report is prepared (the brief also overnight). Otherwise they are written by our rules	AI switched on for personal features
Checklist: steps for a task	Only the one task you choose: its title, its note (which may hold a link you pasted), its due date and its length. Nothing else from your account, and never your other tasks	Only when you tap “Ask your AI instead” (or “Break down” on a task our rules have no steps for), with your own AI connected and AI for personal features on. The steps are a suggestion: nothing is added until you add a step. Without your own AI, steps, dates and the time plan come from our rules on your device	AI switched on for personal features
Timetable scan	The text of your timetable. A photo or PDF is read on your device (the PDF's text layer, or on-device text recognition) and never leaves it; you check the text, and only that text goes to our server, where our rules read it like pasted text - it is not stored and nothing is sent to any AI. Only if our rules cannot read it and you tap “Let my own AI read the text” does that text go to your own AI - after removing the lines labelled with a name, ID number, phone, e-mail or address, and every e-mail address, link, IBAN, phone number and long number	When you scan a photo or PDF or paste a timetable; to your own AI only after you tap “Let my own AI read the text”, with your own AI connected	Your tap on that button
News, markets, Earth explanations	Public data only (headlines, prices, place facts); nothing about you	Only with your own AI connected. Otherwise summaries and “why it matters” come from the publisher's own text by our rules	-

On-device processing: recordings, photos, note photos, scanned PDF pages and timetable photos or PDFs are read only on your device: speech-to-text (Whisper tiny) for audio, printed-text recognition (Tesseract) for images (also photos and scanned pages of a school curriculum document; the image is sharpened on your device first) and a PDF's own text layer where it has one. The audio and images are never sent to our server or to any AI; our server refuses requests that carry them. The model files (about 9 MB for text recognition, about 60 MB for speech) are downloaded only after you agree (the size is shown first; Settings → On-device AI), from our own server (agenticly.live) - which fetches the speech model at a fixed version from Hugging Face (huggingface.co) without sending anything about you - checked with SHA-256 and kept in this browser (Cache Storage) until you delete them. They contain no personal data. After that they work offline. Without them, recordings and photos are not read (except by a browser transcript, if you turned it on - see below). Handwriting is usually not recognised. The text these models produce stays on your device unless you then use a feature that sends text - for example notes AI to your own AI, a clean transcript (“Your connected AI” under Transcripts) or a timetable's text to our server's rules.
Browser transcripts (optional, off by default): when you record, you can let your browser turn speech into text while you record (asked once on your first recording; Settings → Notes & Voice). This uses the browser's own speech recognition: in Chrome and Edge the browser sends the audio to its maker's speech service (Google or Microsoft) under that company's terms to produce the text; Safari may do it on the device. Agenticly does not send the audio anywhere for this and receives only the text, which is saved in your note. Only if you pick “Your connected AI” in the same Settings list (it needs your own AI connected and turns this browser transcript on), the text this draft produces - never the audio - is also passed to your own AI to be written out as a clean transcript (see Your own AI); with plain “Browser transcript” it is not. Leave it off if you'd rather nothing leaves the device. In Meeting mode this includes every voice the microphone hears, other participants’ too (in Chrome and Edge their voices go to that speech service as well); Meeting mode has its own switch for live text and a notice that says so before your first meeting and whenever you turn it on. AI Chat microphone and Listen: the microphone button in AI Chat uses the same browser speech recognition, but only while you have pressed it (the browser asks for microphone permission); in Chrome and Edge the browser sends the audio to its maker's speech service (Google or Microsoft) to produce the text, which then appears in the message box for you to edit. The switch above does not cover it, because it only ever happens when you press that button. The Listen button on an answer reads it aloud with a voice installed in your browser: most voices run on your device, but some (for example Chrome voices named “Google …”) are online voices, and then the browser sends the answer's text to its maker's speech service. Agenticly itself does not send the audio or the text anywhere for this.
We don't use your data to train AI models.

9. Automated and background processing

Shared data - news, markets and exchange rates, weather, pharmacies on duty - is refreshed automatically by scheduled jobs. These jobs use no personal data. The school holiday calendar uses the official MEB dates built into the app (through the 2026-27 school year); later years are shown as labelled estimates, and you can enter your own dates. No web search and no AI is used for it.
Per-account jobs: reminders are checked regularly and send a notification when due; the morning brief is prepared overnight and, only when your own AI is connected and AI for personal features is on, the day's facts go to your own AI for it. They use only data stored with us (your timetable, tasks, reminders, check-ins, saved places, preferences) and never the content of your connected accounts (Gmail, Calendar, Drive).
On your device, while Agenticly is open: new Classroom notifications and coursework are checked and new homework is added; queued note jobs (transcripts, PDF analysis) run once you allowed them - with your own AI running and “Your connected AI” chosen under Transcripts, a recording’s draft is sent to it for the clean transcript right after you stop, and flashcards and questions are made once after a lesson recording has been structured; your data syncs with your account.
These jobs make no automated decisions with legal or similarly significant effects; they only produce suggestions and summaries.

10. Service providers

We don't sell your data or share it for advertising. We use these providers to run the service:

Provider	Purpose	Location
Render	Server hosting and database: every document stored for your account, on the server's disk (encrypted as described above)	Frankfurt, Germany (EU)
Cloudflare	Domain, network and attack protection (traffic passes through)	Global
Google	Sign-in (OpenID), maps/weather/routes; optional fonts; delivery of encrypted reminder notifications to Chrome and Android browsers	US and other
Anthropic	Only if you connect your own Claude: processes the AI jobs your bridge sends from your computer, under your own Claude account and terms	US
OpenAI	Only if you connect your own ChatGPT: processes the AI jobs your bridge sends from your computer, under your own ChatGPT account and terms	US
Ollama (ollama.com)	Only if you choose the local model (Qwen through Ollama): your computer downloads the model files once from ollama.com, which sees your IP address and the model name; the AI work itself runs on your computer and no text is sent anywhere	US
Microsoft	Optional sign-in and connected accounts (on your device); delivery of encrypted reminder notifications to Edge and Windows browsers	US and other
Mozilla	Delivery of encrypted reminder notifications to Firefox (sees the push address, the time and the size of the message, not its content)	US and other
Apple	Delivery of encrypted reminder notifications to Safari (sees the push address, the time and the size of the message, not its content)	US and other
Lemon Squeezy (merchant of record: it sells Formal and Professional to you and handles payment, invoices and taxes)	Checkout, payment, invoices, the customer portal and subscription changes. From us it receives your account id and, if you confirmed it, your email address; it collects your name, payment and billing details itself, under its own terms and privacy policy	US
Resend	Account emails: verification, password reset and security alerts (new-device sign-in, account locked, 2-step verification changes)	US
Open-Meteo, MET Norway, OpenStreetMap-based services (Nominatim, Photon, Overpass, OSRM)	Weather, address and school search, nearby places and routes (coordinates/queries without identity)	EU
TomTom	Optional routes and traffic (coordinates without identity)	EU / other
RainViewer	Weather radar images, loaded directly by your browser (no personal data)	EU / other
Yahoo Finance, Twelve Data, CoinGecko, Google News, and the pharmacy-on-duty services CollectAPI and NosyAPI	Market prices and charts, news, pharmacies on duty. Our server sends them, without your identity, what you type into the watchlist ticker search, the symbols on your watchlist, the country edition and the city you chose for local news, and the province and district for pharmacies on duty	[processing regions of these providers - to be confirmed]

We may also disclose data to authorities where the law requires it.

11. International transfers

Some of these providers are outside Türkiye. Transfers are based on [standard contracts / explicit consent - to be set with a lawyer] under Article 9 of KVKK. Details in the KVKK notice.

12. Retention

Account data while your account is open; deleted within 30 days of account deletion, backups included.
Sessions end when they expire (30 days with “Remember me”, otherwise when the browser closes) or when you sign out.
Daily snapshots of your day's facts (timetable, tasks, check-ins) are kept for 400 days for your trends; daily summaries and AI analyses (mail analyses, reply drafts and weekly ones included) for 60 days; route and news-brief caches for a few days. The readable text of a link our server fetched for you (a saved link, a summary) is cached for 7 days.
Security log (e.g. sign-ins and failed attempts, without content or location): the most recent 300 events per account, while the account exists.
AI chats: until you delete them or your account. Uploaded voice recordings: until you remove them or turn the upload off and remove them. Feature-use counters: 120 days. The AI usage list: at most 48 hours, in memory.
Text sent to your own AI is not stored by us for AI (answers to identical requests stay in server memory for a limited time, see section 8); storage at Anthropic or OpenAI follows the terms of your own account.
Daily briefs (Library → Files): only on your device, for 30 days (see section 6). No generated reports, spreadsheets or presentations are kept on our server. A meeting’s presentation and workbook are made when you tap and are not kept by Agenticly at all (the file you download is yours).
Study buddy links and shared lists: until you unlink, leave or the list is deleted (see section 5, the shared rows); unlinking deletes the shared record at once. Invite codes: at most 24 hours, as one-way hashes. The weekly summary for a parent or teacher is not kept.

13. Security measures

All traffic is encrypted with HTTPS; a strict Content Security Policy applies.
App Lock PINs (and any remaining older passwords) are hashed with Argon2id (PIN hashes are additionally encrypted with AES-256-GCM); saved places are encrypted with AES-256-GCM.
The session cookie is HttpOnly and SameSite; write requests pass an Origin check (CSRF protection).
Sign-in and AI requests are rate-limited. To keep sign-in and PIN locks in place across a server restart, we keep the counters of failed attempts - under a keyed one-way hash of the device address or account, never the address or email itself - for as long as they apply, at most 24 hours.
App Lock: while locked, the server withholds your data (HTTP 423).
You can see every session in Settings and end one or all of them.
Email content, note text, addresses and coordinates are stripped from logs automatically.

14. If we were breached - what could a third party do?

No system is perfect, so here is the worst case, plainly: someone gains unauthorised access to our server or database.

What they could not get

Your Gmail, Calendar, Classroom, Drive and Microsoft access tokens - they are never on our server. An attacker could not read your mail, see your calendar or open your files.
Your mail, calendar and file contents - they never reach our server.
Your App Lock PIN (or an older password) - only an Argon2 hash exists. A short PIN is easier to guess from its hash than a password, which is why a PIN never signs anyone in: it only opens Agenticly on a device that is already signed in with your Google account.
Your saved places in readable form - unreadable as long as the encryption key, kept separately, is safe.

What they could see

Your email address, name and account details such as language and time zone.
Notes, tasks, timetable, check-ins and files you saved with us (fields that are not encrypted).
Your session list (device type, browser).

What we would do

Revoke every session, so everyone has to sign in again.
Rotate encryption keys and session secrets.
Notify the Turkish Personal Data Protection Board within 72 hours and affected users as soon as possible, as KVKK requires.

What you should do

Choose a new App Lock PIN, and secure your Google account (change its password if you used it elsewhere).
Check connected apps in your Google and Microsoft accounts (the connections live on your device anyway, but it's good to be sure).
Watch out for suspicious emails claiming to be from Agenticly - we never ask for your password.
Use a passkey where you can, and turn on App Lock.

15. Your rights

You can access, correct, export and delete your data (Settings → Account & Security). You can withdraw AI permissions at any time. For your rights under Article 11 of KVKK, see the KVKK notice.

16. Changes and contact

If we update this policy, we change the date and announce important changes in the app. Questions: contact@agenticly.live.