Privacy at instanceof.ai
Service operator: Thiago Bajur Alves Miranda.
Your workspace can hold business pages, form responses, bookings, customer conversations and meeting records. This page explains who processes that information and how you control it.
instanceof.ai is operated by instanceof.ai. For the business content, form responses, bookings, customer conversations and meeting records in your workspace you are the controller and we are the processor: we hold the data to run the service for you, and we act on your instructions.
What we collect
| What | Why | Where it comes from |
|---|---|---|
| Your account: email address, name if you give one | To sign you in and attribute what happens in the workspace | You, or Google if you sign in with Google |
| Calendar events: title, time, invitees, join link, organiser | To know which meetings to join, and who may receive a recap | Read-only access to the calendar you connect |
| Google Meet: conference records, participants, recordings and transcript entries | Import calls you select into meeting notes, decisions and commitments, or create a Meet link for a booking | The Google account you connect and the Meet spaces it can access |
| Google Business Profile: locations, opening hours, reviews, replies and posts | Show and update the connected owner's own business presence after review and confirmation | The Business Profiles managed by the Google account you connect |
| Google Drive, Sheets and Docs: files the service creates or that you explicitly open with it | Export meetings and synchronize the forms or bookings you choose | The Google account you connect; the drive.file permission does not provide general access to your Drive |
| Gmail scheduling: messages in the one user label you choose | Turn labelled scheduling requests into booking proposals and replies through your configured booking transport | The Gmail account and label you connect; instanceof.ai never sends through Gmail |
| YouTube: channel identity, uploaded videos, processing state and analytics | Show channel metrics, select the latest embeddable video and upload a recording you choose | The YouTube channel you connect |
| Recordings: audio, and video when you turn it on | To produce the transcript and the notes | The bot, in the calls you send it to |
| Transcripts, notes, minutes, decisions, memory | The product itself | Derived from the recording |
| Usage: recorded minutes, model calls, spoken characters, tool calls | To bill the plan and to operate the service | Measured as the work happens |
| Pages and forms: published content, answers and uploaded attachments | To publish the resources you configure and collect responses | Workspace administrators and respondents |
| Bookings: contact details, selected times and answers | To arrange and manage appointments | Hosts and invitees |
| Webchat: visitor messages, replies and private internal notes | To provide the configured customer support channel | Visitors, authorized agents and workspace users |
| Agent pilot: seat configuration, processing requests and credit receipts | To run authorized tasks and enforce budgets | The authenticated caller and operational metering |
We do not ask for and do not want payment card numbers: those go to our payment processor and we never see them.
Who else processes it
The core service uses the subprocessors listed below. Optional connected applications and Telegram have separate data flows described after this table.
| Subprocessor | What it receives | Purpose |
|---|---|---|
| Oracle Cloud Infrastructure (São Paulo) | Everything in your workspace, at rest | Hosting and storage |
| Deepgram | Meeting audio | Transcription and live transcript |
| OpenAI | Meeting audio, transcript text; when search by meaning is on, the text of your notes, documents, bookings and form answers, and customer messages only if you allow it; your uploaded logo, when OpenAI reads logos for the brand kit | Transcription, meeting notes, search embeddings (text-embedding-3-small) and reading a logo's style |
| Anthropic | Transcript text; your uploaded logo, when Anthropic reads logos for the brand kit | Meeting notes, the in-app agent and reading a logo's style |
| Z.ai | Your uploaded logo, when Z.ai reads logos for the brand kit | Reading a logo's style to suggest page colours and a font |
| Recall.ai (United States) | The meeting link, the notetaker's name and camera image, and the audio and video of the calls it records | Recording a call when our own notetakers are full or cannot join; the recording is deleted at Recall once we have downloaded it |
| TypeSafe AI | The text of a customer message, an inbound email, a calendar event title or a request to the assistant | Sorting what a message asks for, spotting automatic replies and picking the tool for a request or the kind of a meeting |
| Oracle Cloud Email Delivery | Recipient address, meeting title, the summary | Sending sign-in links and recap emails |
| Your email address if you sign in; data from each optional Google app listed above only after you connect it | Sign in and the Calendar, Meet, Business Profile, Drive, Gmail or YouTube feature you choose | |
| Creem, when payment is enabled | Billing identity, selected product and payment information | Payment processing and subscription management as merchant of record |
Google user data
Google integrations are optional, disabled until the operator has the required provider approval, and authorized separately. Connecting one Google app does not grant another app's permissions. instanceof.ai accesses only the data needed for the feature you choose: Calendar events; Meet conference records, participants and transcripts; Business Profile locations, reviews, posts and hours; files created or explicitly opened with the app in Drive; messages in the single Gmail label selected for scheduling; and the connected YouTube channel, videos and analytics.
We use Google user data only to provide or improve the user-facing features you authorize. We do not sell it, use it for advertising, use it to determine creditworthiness, or use it to train generalized AI models. We do not transfer it except to subprocessors that are necessary to provide the feature with your consent, for security, or to comply with law. People do not read it unless you affirmatively ask for help with specific data, access is necessary for security, or the law requires it.
Google integrations can also be connected through a provider bridge while our own Google client is in review. If the operator turns that bridge on and you connect an app through the bridge instead of our own Google client, Composio holds the account connection and relays the same requests to Google. It receives the request and Google's answer for the feature you authorized, and never your recordings or transcripts. The consent screen on the bridge is the provider's, so it asks for what their verified Google client is authorized for rather than the narrower permission we need: for Gmail that is access to the whole mailbox. The Gmail scheduling feature still reads only the single label you select and never writes to your mailbox, on either connection; what your agents may do in Gmail through a connected app is described under Connected apps and agent actions. Connecting the same account with our own client replaces the bridge connection and ends Composio's access to it.
Refresh tokens are encrypted at rest. You can disconnect an integration in the product or revoke it in your Google Account. Disconnecting deletes the active refresh token and prevents future access. Deleting the workspace removes Google-derived data held by instanceof.ai from the live service; deleted data leaves rotating backups within 30 days. Files created in your Drive and videos uploaded to YouTube remain in those Google services until you delete them there, because disconnecting cannot recall a copy already delivered to your account.
instanceof.ai's use and transfer of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.
Business tools and agent access
Published page and form content is intended for its public audience. Form answers, booking administration and internal conversation notes are not public page content. Access is checked against the workspace, caller and applicable resource permissions.
When you connect your own agent, the information returned by its authorized tools leaves the platform for that agent and its providers. Their processing terms apply. A public webchat worker uses separate conversation credentials; these do not grant workspace administration access.
The internal hosted pilot sends the current request and authorized tool results to the configured OpenAI or Anthropic provider. This can include form responses or booking data when requested and authorized, not only meeting transcripts. Browser pilot turns are stateless. Operational request records contain identifiers and status rather than the prompt or answer. Existing Telegram transport retention is described below. Credit receipts are operational records, not customer message content. Uploaded content and public visitor messages are treated as untrusted input.
We do not use your meetings to train models. Our contracts with the model and transcription providers exclude training on the data we send to them. This statement does not establish the terms of optional applications you choose as delivery destinations.
We will publish a notice before adding a subprocessor, so a customer who objects can leave before the change takes effect.
Optional app delivery and Telegram
When you connect an app and approve a hosted delivery, Composio manages that app connection and processes the selected action text, destination, pseudonymous connection owner and meeting source reference. The selected app, currently GitHub, Notion or Slack, receives the delivery. Its access and retention settings apply to copies there. We do not automatically attach meeting audio or the full transcript. Explicit automatic rules authorize the selected content from future matching shared meetings. They are off by default and can be paused.
Connected apps and agent actions. When you connect an app through Composio (Gmail, Google Calendar, Drive, Sheets, Docs, Tasks, Meet, YouTube, LinkedIn, Outlook, Teams, OneDrive, Excel, Slack, Notion, Todoist or GitHub), you choose what your agents may do with that account: read, create and edit, send and publish as you, and delete. All four are selected unless you untick them, and you can change the choice, limit one agent or disconnect at any time. Composio holds the account connection with the widest access the provider offers; each action an agent runs goes through Composio to the app, and its result returns to the agent that asked. We keep a receipt of each action (the action name, the account and the outcome), not its content. Messages, posts and changes made this way stay in the app, and disconnecting cannot recall them.
When you pair a private Telegram chat, Telegram processes your messages and the answers you request. The configured model provider processes requests that need reasoning. A short lived link binds your chat to your workspace identity. Groups and channels cannot pair. The control database holds pairing identities without meeting content. Pending messages are held in the workspace database, cleared after processing, and queue records are removed after seven days. Recent chat context is stored per person in the workspace for up to 30 minutes, limited to three exchanges. Send /new or disconnect to clear it. Meeting access changes or meeting deletion also clear this context. Expired context is excluded immediately and removed by the next cleanup.
Before any account exists, the bot also answers people who write to it. That public conversation reads no workspace: Telegram processes the message, the configured model provider composes the answer from public product information, and the control database holds the chat identifier, the language and up to six recent messages for at most 30 days. Sending /stop ends it, and connecting an account deletes the conversation. A short lived link, valid for 30 minutes and usable once, connects that chat to the workspace of whoever opens it while signed in.
Plans and delivery receipts stay in your workspace and are deleted with the source meeting. Changing access or disconnecting an app prevents future use; it cannot recall a copy already delivered. These optional services require operator configuration and your connection or pairing before use. Their contractual terms and retention settings must be reviewed before activation; no blanket zero retention or no training assurance is made for every downstream destination. The external agent queue remains separate and uses your own agent's credentials.
When you connect an Instagram professional account, Meta sends us the comments, direct messages, story replies, mentions and button taps that arrive on that account, with the sender's Instagram scoped identifier and, when Meta shares it, their public username. We keep them in your workspace to show your inbox, to rank contacts and to run the campaigns you turn on. We also cache the captions of your recent posts, never the images. If you connect through the Composio bridge instead of our Meta app, Composio holds the account connection and relays the same messages and replies. Through either connection, and only for the permissions you leave on for the account, your agents can also publish photos, videos, stories and carousels, read the account's and its posts' insights, send images in conversations and delete comments.
Campaigns send only the replies, links and questions you approved, only to people who wrote to you first, and only within Meta's messaging rules. Links in those messages go through a short address of ours that records the first click; it carries nothing about the person. Values a person types when a campaign asks for them, such as an email or a phone number, are stored on that contact with the text they agreed to. When a campaign uses the hosted model, the configured model provider receives the recent conversation and the facts you wrote for that campaign.
Messages written by the public are treated as untrusted data and never enter meeting memory. Removing our app from the Instagram account revokes the connection. A data deletion request made through Instagram, or erasing a contact in the dashboard, deletes the messages, conversations, links and collected values we hold for it; Meta keeps its own copy of the conversation.
Where it is kept, and for how long
Data is stored in Oracle Cloud in São Paulo, Brazil. Each workspace has its own database and its own file storage; nothing is pooled between customers.
Retention is your setting. By default meeting recordings, transcripts and notes stay until you delete them. Set a retention window and the sweep deletes the recording and the video once they pass it, while the transcript and the notes stay. Deleting a meeting deletes its row, its transcript, its facts and its media. Deleting a workspace erases its database and its files.
Ask conversation history is saved in the workspace for the signed in user or API key. You can clear its contents in the chat. Conversations expire after 90 days without an update and are removed by the next cleanup. Deleting a meeting also deletes the conversation attached to it. Workspace Ask history is separate and can contain information from several meetings.
Meetings can be shared with the workspace or private to their creator, explicitly granted members or API keys, and workspace owners. Existing meetings remain shared unless changed. These permissions also apply to search, memory, derived notes, recordings and follow-up jobs. Private meetings are excluded from native recap delivery, public recap links and webhook payloads. Changing access clears shared derived caches and workspace Ask histories to prevent old cached answers retaining newly private content. It cannot revoke copies already delivered or viewed. Original calendar events retain their calendar's access rules.
Interface language preferences use a first-party cookie. Translations and AnimatedIcons assets are served locally with the application; changing language or playing an icon animation sends no meeting content to an icon or translation provider.
Corrections keep a history of the notes and transcript before and after the edit. That history stays with the meeting and is deleted with it. Booking forms that encounter an error without JavaScript can save a temporary draft on the server. The browser receives only a random token in an HttpOnly cookie. The draft expires after 15 minutes and is removed by the next cleanup.
Backups are kept for 30 days. A record deleted from the live service is removed from backups as they roll off.
Your rights
Under the LGPD and the GDPR you can ask for a copy of your data, a correction, or its deletion, and you can object to processing. In the product:
- Export: Workspace owners can use Settings, then Export workspace. It returns everything as JSON. Other members can export individual meetings they may access.
- Delete a meeting: the meeting page. It removes the media too.
- Delete the workspace: Settings, then the danger zone. Files and rows both go.
- Anything else, including a request on behalf of somebody who was in a meeting: write to hello@instanceof.ai and we answer within 15 days.
You do not need to ask us to delete something you can delete yourself. That is deliberate.
Recording other people
This matters more than the architecture. A notetaker that joins a call and records everyone is subject to wiretap and biometric-privacy law. Several US states, and Brazil in several contexts, require the consent of everyone present; damages are per violation, and there is active litigation against hosted notetakers on exactly this point.
So the bot appears in the participant list under its own name and posts a recording disclosure in the meeting chat by default. Both are on by default and should stay on. That disclosure is a consent artefact, never an advertisement.
You are responsible for having the right to record the meetings you send the bot to. Ask for consent when the jurisdiction or the room calls for it. Some rooms, legal, HR, diligence, medical, should not be recorded at all.
Recap email boundaries
Recap email is off by default. A calendar owner turns it on and chooses nobody, only themselves, or everyone on the invite. Each recipient gets a separate message, so no recipient list is exposed. Declined guests and room resources are excluded. Every message carries one-click unsubscribe, and an opt-out is honoured before any future message is queued. The signed link expires after 30 days and shows the summary only, never audio, video or the transcript.
We do not send marketing email to people who appear in your meetings. They are your contacts, not our leads.
Security
Access to the service needs a session or a workspace API key. Keys are stored hashed and can be revoked at any time. Google refresh tokens are encrypted at rest with AES-GCM. Traffic is TLS only. Requests are rate limited per credential. Outbound URLs you supply, for webhooks or an avatar, are refused when they resolve to private or reserved addresses.
Report a vulnerability to hello@instanceof.ai. We will acknowledge within two working days and will not pursue anyone who reports in good faith and does not access other customers' data.
Contact
hello@instanceof.ai for anything on this page.
instanceof.ai uses YouTube API Services. Your use of YouTube through instanceof.ai is also subject to the Google Privacy Policy. You can remove instanceof.ai access in your Google Account permissions or disconnect YouTube in the product.