Privacy policy

Last updated: 2026-09-24

Twin2.me builds a "twin": an assistant that answers in your voice, from the material you give it. This policy covers what we collect, what for, who we share it with and how long we keep it. The meeting assistant has its own section, because it records conversations with more people in them than you.

Who is responsible

Twin2.me is the data controller. Write to us any time at info@twin2.me, which is also the channel for exercising your rights.

What we collect, and what for

Your account. Name and email, plus your profile photo if you sign in with Google. They come from Google or Microsoft when you sign in with them, or you type them yourself if you sign in with email and password. For the password we store only an scrypt hash, from which the password cannot be recovered. And, to stop anyone guessing passwords, we note how many times each email address and each IP address fails to sign in, and delete that within a day.
The material you upload — files, recordings, links, whatever you connect from your clouds and your apps — to build your twin. It is yours and you can delete it.
Your calendar, if you connect it. See below.
The audio of meetings you send the assistant to. See below.
Minimal telemetry about product use (what failed, how long it took, how many tokens an answer cost). It is not shown to anyone outside the team and is not used to profile you.

We use no analytics or advertising cookies, and the public pages load nothing from a third-party domain: no fonts, no measurement tags, no social widgets. There is no profiling and no automated decision-making about you.

Your calendar

If you connect Google Calendar or Outlook, we do it through OAuth and with a consent separate from signing in: you can use Twin2.me without connecting any calendar.

What we read. Only your primary calendar, and at three moments:
- When you open Calendario in Twin2.me, the events in the range you are looking at (a week, a month or, at most, a year), private ones included: title, time, whether it has a video call, your response and how many guests there are. Only to show them to you on that screen: they are not stored.
- When you ask your twin about your schedule in Hablar, the events in the range you ask for (31 days at most), only for that answer: those are not stored either.
- If your plan includes the meeting assistant, every few minutes, the events for the next 24 hours: title, start and end time, who organises it, who is invited (name and email), and the text where the video-call link usually lives. The assistant skips events marked private and never stores them.
What for. To show you your schedule, so your twin can organise your day and, with the assistant, to know which meeting to join and at what time, and to put a name to who is speaking in the transcript.
What we write. Two things. We add the assistant's email to the guest list of the meetings you mark, and remove it if you change your mind. This is done without notifying the other guests, and we touch nothing else on the event. It is what lets the assistant join without somebody having to admit it by hand. And, only on Google Calendar and only when you press «Hacer» on a card in Hablar, we create on your primary calendar the events you asked your twin to add: with no guests and without notifying anyone.
What we store. For each meeting the assistant is going to join: the title, the time, who is invited, and the join link encrypted — it carries the room password, so it never leaves through our API and never appears in a log. Your calendar tokens are stored encrypted with AES-256-GCM.
Disconnecting. When you disconnect a calendar, meetings that have not started yet are cancelled and the permission is revoked. Whatever already reached your material stays, because it is yours, and you can delete it from there.

Twin2.me's use of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. Specifically: we do not sell that data, do not use it for advertising, do not use it to train models, and no human reads it unless you ask us to in order to resolve an issue or the law requires it.

The meeting assistant

If your plan includes it and you switch it on for a specific meeting, a twinme-branded assistant joins that video call, records the audio and leaves the transcript in your material. Here is what it does and what it does not:

It announces itself. It joins under a name that says what it is, and posts a message in the meeting chat saying that it is recording and for whom. On Zoom it also triggers the platform's own recording indicator, which cannot be suppressed.
Audio only. It does not record video, does not record screen sharing, and does not speak.
The audio is deleted. It is transcribed and, as soon as the transcript is ready, the audio file is erased: what remains is the text. We do not keep both.
Other people in the room are protected. The transcript enters your material as a private source; what other people said does not reach the search index or your twin's answers except after a leak filter and a quarantine that you resolve, not an operator of ours. Third parties' full names are blocked.
The minutes are for you. From the transcript we draw a summary, what was left to do —and who took it on— and what mattered, even if you were not there. You see them in Calendar and we email them to you, which you can turn off in Settings. Nobody else receives them, they do not feed what your twin tells other people, and removing the meeting from your material removes them.
Your twin does your tasks. If the minutes say you have to do something that is done in an app you connected —sending an email, scheduling, creating a document—, your twin does it in your name with that app and tells you in Twin2.me. Anything that moves money, and anything past a few actions per task, per meeting and per day, is left for you to approve. It works from what was said in the meeting, including what other people said. You can turn it off in Settings.
What we do not control. The organisation hosting the meeting can stop the assistant from joining. That is not a failure: it is their decision, and we respect it.

Recording a conversation is your responsibility. In many places the law requires notifying everyone on the call, and in some it requires their express consent. The assistant gives notice on its own — its name and a chat message — but that does not replace you: you are the one who switches recording on, and you are the one answerable for having whatever permission your jurisdiction requires. If someone in the room objects, ask the host to remove the assistant and delete the meeting from your material.

Who we share it with

We do not sell or rent your data, and we do not use it for advertising. To operate we rely on these processors, and on nothing else:

Amazon Web Services (United States): the server, the database, file storage, the emails we send you and the meeting assistant.
Deepgram and Groq: transcribe the audio. They receive the audio file and return the text. Meeting audio goes to Deepgram only.
Language-model providers — today DeepSeek and Groq — to distil your material and answer. If you connect your own AI account, traffic goes to yours and not to ours.
OpenAI: lets your search index match by meaning and not only by words. It receives the text that enters the index and the questions people ask your twin, and returns numbers. It never receives your original files.
Composio: connects the apps you link, such as Gmail, Google Docs or Notion. It holds the permission for each one, and everything we touch in them goes through it: what we bring into your material, and what your twin reads or does when you ask it to in Hablar — anything it writes, only after you approve it. It does not store the content of those calls.
Google Workspace: our email, and the account the assistant uses to join Google Meet. When you send it to a Meet meeting, the invitation — title, time, guests and link — reaches that account.

None of them trains models on your material. If that ever changed, this page would change first.

How long

Your material, while you have an account or until you delete it. Deleting a source deletes it for real, file included.
A meeting's audio, until the transcript is ready. After that it does not exist.
Calendar events that never got recorded are cleaned up with the rest of the meeting history.
Your account, until you ask us to delete it. Then everything hanging off it goes.

How we protect it

No model is trained on your data. A twin has no weights: it is your corpus, your distilled voice and your persona specification on top of a model that does not change. There is no model trained on you.
Your original files are not indexed and a query does not touch them. Only what is marked publishable enters the search index.
Your cloud and calendar permissions are stored encrypted with AES-256-GCM, and so are meeting links, because they carry the room password.
Your files do not pass through our server: they upload straight to storage with a signed, single-use permission.

Your rights

You can know, update, rectify and delete your data, withdraw your authorisation and request proof of it, and complain to the Superintendencia de Industria y Comercio (Colombian Law 1581 of 2012 and Decree 1074 of 2015). If you are in the European Union you also have the rights of access, rectification, erasure, restriction, portability and objection (GDPR articles 15 to 22) and to complain to your supervisory authority. Write to info@twin2.me and we will answer.