Skip to content
OpenScribe Docs
Privacy policy

OpenScribe privacy policy

Last updated: November 22, 2026

This policy covers OpenScribe at https://scribe.akyonlabs.com, operated by Akyon Labs. It explains what OpenScribe collects when you sign in, connect Google Calendar, Outlook or Zoom, connect Outlook mail (or Gmail on a self-hosted server), record meetings with the Mac app or Chrome extension, and use the workspace, and who receives it. It adds product details to the Akyon Labs privacy policy.

TopicIn short
Sign-inGoogle sign-in with openid, email and profile, limited to the server's allowlists
Google dataEach Google scope, what it reads or changes, who receives it and how it is protected
CalendarOptional Google, Outlook and Zoom access, with separate Google consent for invitations
EmailOptional read-only Outlook, and Gmail on self-hosted servers, matched to existing people
ZoomOptional connection to create scheduled calls
MeetingsRecorded on your Mac first, then synced to your workspace
AI and transcriptionBounded data goes to the providers your workspace uses
AnalyticsAnonymous product events, no content, can be turned off
DeletionDisconnect, delete items, or ask us to delete your account

OpenScribe does not sell your data or Google user data, and does not use it for advertising.

Account sign-in and access#

OpenScribe is a private workspace where each user sees only their own records.

  • Google sign-in scopes: signing in to the web app or the Mac app (POST /api/collections/users/auth-with-oauth2) requests only openid, email and profile
  • Allowlist: only verified Google accounts on the server's sign-in allowlist (OPENSCRIBE_USER_EMAIL) or admin list (OPENSCRIBE_ADMIN_EMAIL) can sign in or create an account. Only admins can change AI provider, model and agent settings. Removing an address blocks sign-in but does not delete the existing account record
  • Self-hosted sign-in: an operator can enable email and password for verified accounts on the instance's allowlist. The browser sends the password to that instance for authentication. PocketBase stores a password hash, and the Mac app receives a session token through browser pairing
  • Stored account data: user ID, email address, display name and avatar URL, in the users collection
  • Time format: your Auto, 12-hour or 24-hour choice is saved with your account. Auto chooses AM/PM when a server behind Cloudflare receives a US country header, and 24-hour time otherwise. The hosted service is not behind Cloudflare, so Auto uses 24-hour time there. No country or IP is stored for this preference and no external geolocation request is made
  • The Mac stores whether you completed optional setup and whether you requested screen recording access in local preferences.
  • Mac sign-in: the Mac app saves its session in the macOS login Keychain (com.akyonlabs.openscribe) by default. Turn off Remember sign-in on this Mac to keep it in memory only. Older builds turned it off on every sign-out, so the first launch after updating turns it back on once. The browser hands the app a one-time sign-in grant over 127.0.0.1, and the server stores only its hash until the app uses it
  • Mac updates: release builds download appcast.xml from the fcakyon/openscribe-releases GitHub releases every four hours, and the new DMG when one exists. GitHub sees your IP address and the app version in the user agent. No account or meeting data is sent

Scheduled meetings store your supplied title, planned times, timezone, call link and participant email addresses. Participant entry can suggest saved workspace people. A meeting without guests, without a connected Google calendar or saved with Save without inviting sends no invitations.

When you connect a provider while scheduling, the browser keeps the unfinished form in that tab for your OpenScribe account.

Google user data#

This section covers data from Google APIs. Google sign-in, Calendar, Meet links and invitations use one Google OAuth client, and each optional scope is requested only when you start the matching feature.

ScopeRequested whenData OpenScribe readsUsed forChanges in your Google account
openid, email, profileSign-in and every Google connectionGoogle account ID, email, name and avatar URLSigning you in and labeling connected accountsNone
calendar.events.owned.readonlyYou connect Google CalendarEvents on calendars you own: ID, status, title, start and end, description, location, join links, attendees and organizerCalendar views, meeting series, links from recurring event descriptions, matching attendees to your people, meeting contextNone. Read only
meetings.space.createdYou choose Google Meet while schedulingThe Meet space OpenScribe creates and its linkAdding the link to your scheduled meetingCreates one Meet space per request. No Calendar event or invitation
calendar.events.ownedYou save a meeting with guests from a connected Google calendarThe event OpenScribe createsInviting the guests you addedCreates one event in your connected calendar, and Google emails your guests. Other events are never edited or deleted
gmail.readonlyYou add a Gmail mailbox in Emails (self-hosted servers with Gmail enabled; off on scribe.akyonlabs.com)See EmailShowing mail with people already in your workspaceNone. Read only

Accounts connected before OpenScribe switched to calendar.events.owned.readonly may still hold the broader calendar.readonly or calendar.events.readonly grant until you revoke it at Google. OpenScribe uses it only for the same event reads.

Who receives Google data#

RecipientGoogle-derived dataWhy
Meeting intelligence: OpenAI, Anthropic or Google Gemini, as configuredMeeting titles, attendee names and emails, links taken from event descriptionsSummaries, action items, Q&A, tags, people and project updates you use
Decision provider: OpenAI, Google Gemini or Typesafe, as configuredMeeting titles, attendee names and emailsDuplicate and series checks after extraction
Typesafe System One, when TYPESAFE_API_KEY is setAttendee names and emails as people, your connected Google account email, name and calendar IDSearch suggestions
MCP clients you authorizeMeeting records with titles, participants, links and Meet linksThe client requests you approve
Google CalendarTitle, schedule, timezone, call link and guest emails you enterInviting the guests you added

No full event description, attendee response status or Meet space name is sent to an AI provider. Product analytics record only that a Google calendar was connected, never its content. Each AI provider processes data under the workspace administrator's API account and that provider's API data terms. An MCP client follows its own terms and your settings with that client.

Data protection#

  • In transit: the hosted service serves HTTPS and redirects HTTP to HTTPS
  • Provider tokens: Google and other provider tokens are encrypted with AES-256-GCM before storage and are never returned by any API
  • Access rules: workspace records are readable only by their owner through the API. AI provider settings and agent configurations are shared server settings that only admins can change. Admin rights give no access to other users' records in the app
  • Operator access: the server operator administers the database with PocketBase superuser access and holds the token key
  • Logs: the server runs without database statement logging
  • Storage: the hosted database runs on one Akyon Labs server through OpenShip. Content other than tokens, such as meeting titles, attendee emails and transcripts, is protected by the access rules above and is not separately encrypted

Google data retention#

OpenScribe reads calendar events live for each view or sync and keeps no copy of your calendar. Disconnecting the calendar deletes its meeting series, the links taken from event descriptions and its hidden-event choices. Meetings, people, Meet links and invitation IDs stay until you delete the meeting or have your account deleted. See Retention, disconnection and deletion.

Calendar#

Connecting a calendar is optional. Choose Google Calendar, Outlook Calendar or Zoom in Calendar. Outlook uses delegated Calendars.Read separately from Outlook mail. The Zoom connection creates meeting links and also reads the native calendar when you grant optional calendar access. OpenScribe reads the primary calendar for Outlook and Zoom.

Calendar access and invitations#

  • OpenScribe requests https://www.googleapis.com/auth/calendar.events.owned.readonly with openid, email and profile, using the OAuth 2.0 authorization code flow with PKCE (S256)
  • It lists events from your primary calendar by default, or the calendar you connected. It does not read your calendar list or calendar settings
  • When you save a meeting with guests from a connected Google calendar, OpenScribe requests calendar.events.owned separately. It shares the meeting title, schedule, timezone, optional call link and guest emails with Google Calendar. Google creates the event and emails the guests. OpenScribe keeps the event ID and iCalendar UID to link it back to the meeting

Calendar data used and stored#

A sync (GET /api/openscribe/calendar/events?sync=true) reads a window from 7 days back to 14 days ahead by default.

Only regular events become meeting context. Working locations, focus time, out-of-office blocks, birthdays and Gmail-generated events are excluded. Hiding an event or series stores its calendar account, calendar ID, event or recurring ID, scope, title, start time and account email in your workspace. Hidden events remain available in the searchable Settings manager until you restore them or disconnect that calendar. Hiding never changes Google Calendar or deletes saved meeting notes.

DataStored as
Connected account: provider, provider account ID, email, name, calendar ID, granted scopes, sync status, last sync timecalendar_accounts row
OAuth access and refresh tokensEncrypted with AES-256-GCM and a random 12-byte nonce (encrypted_token), keyed by OPENSCRIBE_TOKEN_ENCRYPTION_KEY or a SHA-256 key derived from the server's secret configuration. No API returns plaintext tokens
Event ID, status, title, start and end, join links (hangoutLink, htmlLink, conferenceData.entryPoints), attendees (email, name, response, self)Shown in your calendar views. Cancelled events are skipped
Recurring series ID and title (recurringEventId)meeting_series, so repeated meetings share context
Non-conferencing links in recurring event descriptionsLink notes in meeting_notes (provenance = "calendar_description", ingested_status = "url_only"). Only the URL is stored
Attendee emails and namesYour people directory (people_profiles, knowledge_entities), matched by email only

Creating a Google Meet link is optional and separately requests meetings.space.created. It creates a meeting space through Google Meet with the selected connected account. Meet link creation alone sends no invitation. The returned link is stored with the scheduled meeting.

Email#

Gmail and Outlook are optional connections separate from sign-in and Calendar. Gmail uses gmail.readonly and is turned off on scribe.akyonlabs.com (OPENSCRIBE_GMAIL=off) pending its own restricted-scope verification; self-hosted servers offer it unless OPENSCRIBE_GMAIL=off. Outlook uses Microsoft Graph delegated Mail.Read and User.Read with offline_access. A Gmail mailbox connected earlier stops syncing on a server without Gmail, and its imported mail stays until you disconnect it. These connections read messages and identify the connected mailbox.

  • Selection: sync searches for mail whose sender or recipient matches an address already attached to another person in your workspace. Your own addresses and addresses shared by multiple people are excluded from that search. Mail unrelated to those people is not imported. Actual message headers are checked before retaining a message, rather than treating quoted or forwarded text as an address match
  • Stored data: encrypted account tokens, search checkpoints with contact addresses and person IDs, message and thread IDs, subject, sender and recipients, timestamps, plaintext message bodies and person/project associations stay in the workspace database. While matching a long thread, OpenScribe temporarily keeps the current redacted section, input fingerprint, progress and provisional project IDs with the thread. Completing classification or changing the message content clears that checkpoint. Attachments are not imported
  • Person matching: address matching happens on the OpenScribe server and does not call an AI provider
  • Project matching: a private CPU service beside the server masks detected financial details (such as bank accounts, IBANs, payment cards, tax IDs and salary or compensation lines), credentials/secrets and government IDs in email and candidate project context, while keeping person names and normal discussion context, before the configured decisions provider receives that context and opaque project IDs. The server reuses redacted project text in a bounded in-memory cache scoped to the workspace. Model detection can miss sensitive details, so this does not guarantee complete removal. A redaction failure prevents the external classification request
  • Display: the Emails workspace and person and project Emails tabs show retained threads and plaintext messages. Opening the original takes you to Gmail or Outlook
  • Sync progress: the server checks connected mailboxes every five minutes while it is running, including when the web app is closed. It searches recent mail for known addresses and up to 90 days for newly eligible addresses. During sync the server keeps address and message counts in memory so the mailbox owner can follow progress. The most recent run result expires after ten minutes. Mailbox sync failure logs record the provider, upstream HTTP status when available, whether Gmail rate-limited the request, error type and saved message count
  • Disconnect: disconnecting an email account removes its local tokens and imported threads and stops further sync. Your original messages stay with the provider

Zoom#

During a Teams browser recording, OpenScribe requests Microsoft Teams live captions. It saves finalized remote text, the Teams participant ID and display name, the result ID used to identify the line, and the spoken language when supplied. Microsoft produces these captions, including when your local microphone uses a device-local model. OpenScribe keeps unmapped Teams receiver audio on your Mac.

When you join a Zoom web call in Chrome, the extension sends its title, room number and presence timestamps to the Mac for meeting notes and calendar matching. The room link omits the passcode and query string. Starting Zoom browser transcription reads new caption text, the speaker’s meeting-scoped Zoom ID and display name, and the caption message ID. The Mac saves those lines locally and syncs them to the workspace. Times mark when OpenScribe observed the caption. This path sends no audio to an ASR provider.

Connecting Zoom stores your Zoom user ID, email, display name and encrypted credentials. Creating a call sends the title and schedule to Zoom and stores the participant join link. Disconnecting Zoom clears the saved connection and its link request receipts. Existing Zoom meetings stay in Zoom.

Meeting capture and storage#

  • Automatic recording: with Start recording automatically on join on, which is the default, the Mac app starts recording without a click when it detects that you joined a Zoom or Teams desktop call, or a Meet or Teams call in Chrome that the extension reports joined. Recording saves audio on your Mac and, when you are signed in, sends it for live transcription as described below. Stop, or closing the Meeting detected prompt, keeps that call from recording again by itself. Turn the setting off in Mac Settings → Recording. Zoom in Chrome still starts only when you choose Start Transcription
  • Live questions: a question you type in the Mac app's Ask, or a quick action such as Catch Me Up, goes to your workspace only when you send it. The workspace saves it with the meeting and sends it to the meeting intelligence provider with this meeting's context listed under Meeting intelligence below and the earlier questions and answers of its Ask thread, the same way as Ask on the web. No earlier meetings are included
  • Notes document: the notes you type on the Mac meeting page are saved in that meeting's local folder as you type and, when you are signed in, sync to the meeting as one note that each later edit replaces. Clearing the notes removes that note from your workspace
  • Enhance Notes: choosing Enhance Notes first syncs your notes, then sends this meeting's context listed under Meeting intelligence below to the configured provider. The summary and action items it returns are saved to the meeting, the same as summarizing it on the web. Your notes text stays as you wrote it
  • Local first: the Mac app writes audio (your microphone, call audio and per-participant streams where available), packet timing, transcripts, notes, attachments and screenshots to ~/Library/Application Support/OpenScribe/recordings/<sessionId>/ on your Mac
  • Your participant: when a platform supplies your local participant identity, local microphone audio is stored under that participant rather than a separate microphone participant. Browser Teams supplies its local signaling ID
  • Workspace sync: when you are signed in, meeting details, participants, transcript segments, notes, uploaded files and screenshots, and extracted knowledge (people, projects, details, evidence links, tasks, tags and suggestions) sync to your own records in the OpenScribe database
  • Meeting state: the Mac syncs each capture session's identifier, start, state and periodic presence updates. Calendar occurrence identifiers link confirmed rejoins to one meeting. Interrupted sessions retain an unknown end time. Session identifiers also keep temporary participant IDs separate across rejoins
  • Duplicate repair: a workspace operator can combine confirmed captures of one occurrence. Their transcripts, notes, files and citations keep their record IDs, and old meeting IDs remain as links to the combined meeting. Conflicting manual summaries remain available as notes
  • Imports: new and previously imported Tactiq and tl;dv meetings are processed in the background. OpenScribe groups copies within your workspace and sends selected transcript passages, meeting context and source speaker labels to your configured meeting intelligence provider to update Mind. The original transcripts and their citations stay available. Connected sources check daily for new meetings. The server uses your encrypted tl;dv API key. The Chrome companion reads Tactiq exports while signed into the bound source account. OpenScribe stores the source account address, last and next sync times and progress. Disconnecting stops discovery and deletes the saved connection, including any tl;dv key
  • Long transcripts: when a provider reaches its output limit, OpenScribe can send smaller sections of the same transcript to the configured provider to finish the meeting update
  • Project matching and merges: meeting extraction receives active project IDs, names and taglines as reference context. Absorbed project names remain stored as aliases. Merges rewrite project links and combine repeated facts and tasks while retaining their source evidence and manual task edits.
  • People matching: in recorded and imported meetings, a participant merges with an existing person by email, or by an exact full name of two or more words, including a new email joining a same-name person without one. A participant without email under your full name becomes you. One-word or shared names stay separate and may appear as a suggestion. No AI call decides a merge. The rules are in People, projects and tasks
  • Financial data guardrail: code-level checks (CheckNoFinancialData, sanitizePersonMetadata) block storing salary, income, tax, bank statement or contract rate fields in people and knowledge records

Automated provider checks send a fixed synthetic speech clip, never your meeting audio. Their private QA reports contain the provider, model, word count, elapsed time and success or failure. A dedicated probe identity has no access to meeting or admin routes.

When you use live transcription, meeting AI or search, OpenScribe sends bounded data to the services your workspace is configured to use. Enabled workflows can also run automatically after a meeting ends.

ServiceReceivesProviders
Speech to textLive or recorded meeting audio: your microphone and remote Meet audio. Meet may send an unresolved transport stream and discard its unattributed text. Mixed Teams browser audio and mixed audio from the Zoom and Teams desktop apps stay on your MacElevenLabs (scribe_v3_live, scribe_v2_realtime), OpenAI (gpt-4o-mini-transcribe, gpt-4o-transcribe, gpt-live-transcribe), Google Gemini (gemini-3.5-transcribe-live, gemini-2.5-flash-native-audio-latest), or a configured OpenAI-compatible endpoint
Meeting intelligence: summaries, action items, Q&A, tags, people and project updatesMeeting titles, participant names and emails, existing owner project names and taglines, transcript segments, notes, attachment and OCR text, selected screenshots, series summaries, and for Q&A your question, the earlier questions and answers of that Ask thread and the transcripts of the earlier series meetings you includeOpenAI, Anthropic or Google Gemini, as configured
Task and deadline resolution (task_resolution)Open task or active deadline text, project and assignee labels, post-creation canonical meeting transcript/note excerpts and post-creation OPENSCRIBE_PII_URL-redacted email excerpts (never unredacted or pre-creation email bodies or restricted financial fields)Anthropic (claude-haiku-5-5 by default), OpenAI or Google Gemini, as configured in task_resolution
Search suggestions, only when TYPESAFE_API_KEY is setYour search text and up to 30 candidate labels and subtitles: names, emails and short briefs of people, project names and briefs, tag names and descriptions, connected calendar emails and names. Never transcripts or audioTypesafe System One (https://api.typesafe.ai/v1/systemone, model jev-1.14.0)
Duplicate and series checks, after a meeting's knowledge extractionExtracted or updated saved project names with stored project names and aliases, task texts and person names with the stored ones they may repeat (stored people's emails where present), and meeting titles with participant names for series checks. Never transcripts, audio or screenshotsThe decision API the administrator selected: OpenAI (gpt-6-luna, the default), Google Gemini (gemini-decision-1) or Typesafe System One (jev-1.14.0)
MCP clients you authorizeThe data allowed by the scopes you approve, until the grant expires or you revoke itClients such as Claude, ChatGPT, Codex or Cursor, through OpenScribe's OAuth 2.1 and PKCE flow at /mcp

Each provider processes this data under the workspace administrator's account and that provider's API data terms.

Mac release notarization#

The release command uploads the built application package to Apple for notarization. It packages the app bundle, not local recordings, preferences or saved sign-in credentials. Signing and notarization credentials stay in the release operator's Keychain.

Product usage analytics#

  • On by default: the web app, Chrome extension and Mac app send a few anonymous product events (for example "recording started" or "import finished with 12 meetings created") to Google Analytics 4 through the OpenScribe server. Turn it off for yourself in Settings → Share anonymous usage data, or for a whole server with OPENSCRIBE_TELEMETRY=off
  • Where events go: the hosted server passes your IP address to Google Analytics with each event. A self-hosted server relays events to scribe.akyonlabs.com without your IP address
  • What an event holds: the event name, the app (web, extension or macos) and its version, whether the server is hosted by Akyon Labs or self-hosted, and a few fixed values or counts. Never meeting titles, transcript text, participant names, emails, URLs or file names. The server keeps one allowlist (server/analytics-policy.json) and drops everything outside it
  • No analytics script: no OpenScribe page loads a third-party analytics script, and the web app sets no analytics cookie. A browser is identified by a random value stored in that browser, or by an earlier Google Analytics cookie if one is still present

Google API Services User Data Policy (Limited Use)#

OpenScribe's use and transfer to any other app of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.

The use of raw or derived user data received from Workspace APIs will adhere to the Google User Data Policy, including the Limited Use requirements.

  • calendar.events.owned.readonly is used only to show your upcoming and recent meetings on calendars you own, link recurring series, surface event description links, match attendees to your workspace records, and support meeting preparation and intelligence you request
  • meetings.space.created is used only to create a Meet space when you ask for a Meet link while scheduling. OpenScribe stores the returned link and does not read or change other Meet spaces
  • calendar.events.owned is used only to create the one event that invites the guests you added. OpenScribe keeps its event ID and iCalendar UID to link it to the meeting
  • gmail.readonly (off on scribe.akyonlabs.com; optional on self-hosted servers) is used only to show retained mail with people already in your workspace
  • OpenScribe transfers Google user data to third parties only as needed for these features (the AI and search providers in Who receives Google data), to comply with the law, or with your explicit authorization (such as an MCP grant you approve)
  • People do not read Google user data unless you agree for specific records, it is needed for security, it is required by law, or it is aggregated and anonymized for internal operations
  • OpenScribe does not use Google user data, raw or derived, to develop, improve or train generalized or non-personalized AI or ML models
  • The hosted service sends Google user data only to AI providers whose API terms do not let them train models on it: the OpenAI and Anthropic APIs, the paid Gemini API tier and Typesafe
  • OpenScribe never sells Google user data and never uses or transfers it for advertising, creditworthiness checks or data brokerage

Retention, disconnection and deletion#

To removeDo thisWhat happens
A Google Calendar accountDisconnect it under Calendar or Settings (DELETE /api/openscribe/calendar/accounts/{id})The account row, its encrypted tokens, its linked meeting_series, the links taken from their event descriptions and its hidden-event choices are deleted at once, and syncing stops. Meetings and people already created stay until you delete them. Disconnecting does not revoke the grant at Google
OpenScribe's Google accessRevoke it at Google Account permissionsGoogle stops issuing tokens. Records already built from Google data stay until you delete them or ask for account deletion
Notes, files, tags, knowledge details, tasksDelete them in the web app (DELETE on /notes/{noteId}, /attachments/{attachmentId}, /tags/{id}, /knowledge/details/{detailId}, /knowledge/tasks/{taskId} under /api/openscribe)The item is removed. A task or detail taken from a meeting stays as a hidden removal record, so later extraction does not bring it back
MCP grantsRevoke them in the web app (DELETE /api/openscribe/oauth/grants/{id})Revocation blocks access immediately. The server stores token hashes, a per-grant rotation secret and the last-use time. Hourly cleanup removes grants more than seven days past revocation or expiry
An ended or failed meetingOpen the meeting, choose More actions, then Delete meeting, and confirmDeletes the synced meeting and its linked records. Local Mac recording folders remain separate
Imported meetingsUndo the import in the web app (DELETE /api/openscribe/import/{source}/meetings)Source transcripts are deleted. The workspace keeps each removed provider ID so daily discovery skips it. Other sources remain available, and a connected source can still import new meetings
A local recordingDelete its folder under ~/Library/Application Support/OpenScribe/recordings/The Mac copy is gone
Saved Mac sign-inSign out in the Mac app, or remove com.akyonlabs.openscribe in Keychain AccessThe saved session is gone
Synced meetings, people or your whole accountAsk your workspace administrator, or email fatihcagatayakyon@gmail.comRecords are kept until explicitly deleted. Deleting an account removes every record it owns, including meetings, people, connected accounts, imported email and feedback records. Removing sign-in access alone does not erase them. Server backups and data already processed by providers follow their own retention

Contact#

For questions about this policy, Google data access or deletion requests:

Mind feedback and persona data#

OpenScribe stores your selected Mind persona, custom node and section labels, agent execution traces and feedback records in your workspace. The optional medical-visits persona can label health-related records that you choose to enter. It does not enable mobile recording or a clinically validated workflow.

Feedback records include the item before your action, the chosen result, relevant entity snapshots and citations, the agent pipeline and version, and a numeric reward. These records can retain earlier text after an individual item is edited, removed or superseded, including evidence from its source meeting. They remain scoped to your account and are not sent as product analytics. They are deleted with the account. Feedback records can contain Google-derived meeting titles and attendee names. No model training or external dataset upload is started by feedback or the training-readiness count, and OpenScribe does not use these records to train any shared or generalized model.

View as Markdown