Trust & data
The part your security team will ask about.
Written for the person who has to forward it. Every promise here names the thing that makes it true, so the engineer you send it to can check rather than take our word — and the last section names what we cannot promise yet.
Isolation
How one workspace is kept away from another.
The separation is enforced by the database rather than by our code remembering to ask politely, and it is re-checked every time we ship.
Your workspace is fenced off inside the database
Not by our code remembering to filter every query — by Postgres itself. The application connects with an account that has no power to look past the fence, so a query that forgets whose data it is asking for comes back empty rather than coming back with somebody else’s meeting.
And it cannot be forgotten later
Isolation usually breaks a year in, when somebody adds a feature and misses a step. Ours is checked automatically every time we ship: if anything that stores your data arrives without its fence, the release fails and never reaches you. It is not a habit we rely on people keeping.
Proven on every build, not asserted in a document
The separation is tested against a real database every time we ship, including the case that matters: a query written without any mention of a workspace, which comes back with nothing. If those tests cannot run, the build fails rather than passing quietly.
Connected accounts unlock for one workspace only
Anything you connect is encrypted with a key belonging to your workspace alone, and your workspace’s identity is sealed into the encrypted value itself. Lift that value into another workspace and it does not decrypt into the wrong hands — it refuses to decrypt at all.
The files it writes stay on our own machines
Summaries, notes and documents the agent produces sit in storage we run ourselves, not in somebody else’s bucket. Every file is filed under your workspace, and nothing in the system can ask for one without saying which workspace it belongs to.
Subprocessors
Who else touches a meeting.
An agent that listens and answers out loud cannot be a closed box — the audio has to reach something that can hear it. So rather than claim nothing ever leaves, here is the whole list, and what each one gets.
Attendee
Runs the bot that joins the call. Receives the meeting link and the call’s audio, and keeps the original recording.
Deepgram
Turns speech into text. Receives the meeting audio.
OpenAI
The voice that answers, and the reasoning behind it. Receives the live audio and the transcript text it works from.
Composio
Carries out the actions you ask for in your connected tools, and holds the sign-in for them. Receives the instruction and whatever comes back.
Arcade
The same, for tools connected before we moved to Composio. Those connections keep working and keep using it.
Mailgun
Delivers mail — the calendar invites that reach the agent, and your sign-in links. No meeting content.
Stripe
Takes payments. Sees who is paying, never anything from a meeting.
Any tool server you add yourself
If you connect one of your own, it receives whatever you point it at. That one is your choice rather than ours.
What we do not do
Four things that are never going to happen to your meetings.
We do not sell your data
Not to advertisers, not to data brokers, not repackaged as anonymised industry insight. Rango is a paid product, which is the whole reason your meetings do not have to be the product.
This page is not watching you
No analytics, no advertising pixels, not even a third-party font loading from someone else’s server. Open your browser’s network tab and count the connections to anyone but us: there are none. A company that sold data would have started collecting it here.
We do not hold the keys to your other accounts
When you connect a tool, the sign-in that lets it act stays in our tooling provider’s vault rather than ours. What we hold is our own key to that provider. Someone who took our entire database would have keys to our own tooling, not to your inbox.
We do not keep the recording
Nothing in our system ever asks for the audio file. It stays with Attendee, who ran the bot that joined your call — so we cannot lose, leak or hand over a copy we never took.
Not claimed
What we do not claim yet.
Everything above names a mechanism you can check. These are the sentences we would like to be able to write and cannot — listed here because you would find out anyway, and finding out from us is worth more than the sentences would have been.
We do not claim your meetings train nobody’s model
The service that turns speech into text improves its models from customer audio unless you opt out, and we have not set that flag yet. Until we do, the honest position is that we cannot promise it — so we do not, rather than quoting the one provider whose default is better and letting you assume the rest.
We do not promise a deletion window
Meeting records are swept on a schedule, but what the agent learned from a meeting is summarised into its knowledge base and that copy outlives the meeting. There is no delete button for it today. "Gone after N days" would be false about the store that matters most, so we will not print it until it is true.
We are not certified against anything
No SOC 2, no HIPAA, no GDPR compliance claim. What is on this page is the mechanism itself, described so your own reviewer can check it — which is less than a certificate and more than a logo.
We say prompt injection is mitigated, never prevented
An agent that reads your documents and acts in your tools can be told what to do by the things it reads. We work to limit what that can reach, and we use the word "mitigated" everywhere on purpose. Anyone selling you prevention is selling you a word.
Twenty seconds from now, you could have a teammate.
Claim your Rango address and bring it to your next call.
Free to try on your first meeting. No card.