What privacy-first should mean for a live AI assistant
Sergei Skrylkov, Founder, Trippi · 2026-09-07 · 6 min read · facts checked against the product on 2026-09-20
“Privacy-first” appears on the homepage of nearly every meeting tool, including ones that record everything you say and keep it indefinitely. The phrase means nothing on its own. These five questions do have answers, and the answers differ sharply between products — including in ways that are inconvenient for the product answering.
1. Where does the audio physically go?
There are three common shapes. Audio can go from your machine straight to a speech provider; it can pass through the vendor’s servers on the way; or it can be recorded and processed in batches. They look identical in the interface and are completely different in what is exposed and to whom.
Ask specifically whether the audio touches the vendor’s own infrastructure, and whether it is written to disk anywhere. “Processed securely” is not an answer to either question.
2. What is kept after the call, and for how long?
This is where slogans usually break. Keeping nothing is genuinely possible; so is keeping the text for a fixed window for a stated reason. What is not defensible is being vague, or saying “nothing is stored” while a debugging table quietly fills up.
We changed sides on this one ourselves, and it is worth saying plainly: Trippi Cue used to keep nothing, and since 18 September 2026 it keeps the recognised text of the call and the answers for 30 days, because a poor answer and a poor question are indistinguishable without them. The policy changed in the same release as the code, which is the part that actually matters.
3. Who else receives it?
Every live assistant has at least two subprocessors: a speech-to-text provider and a language model provider. Both should be named, not described as “industry-leading partners”. If a tool cannot tell you which model provider sees your meeting, it cannot tell you where your meeting went.
4. Does anything join the call?
A bot in the participant list is visible to everybody and usually carries a recording notice. An extension inside your tab is visible to nobody — until you share that tab, at which point it is visible to everybody. Both facts should be stated by the vendor rather than discovered by the user mid-call.
5. What is the training answer?
“We do not train on your data” should cover the vendor and its subprocessors, and should be checkable against what the subprocessors’ own terms say. A vendor that has not read those terms closely enough to answer is telling you something useful anyway.
Why this matters more for live tools
A note taker handles what was said in a meeting. A live assistant handles the same thing plus whatever context you gave it — for an interview, that context is often your own background, written by you, minutes before. The five answers above are the difference between a tool you can use in a conversation that matters and one you cannot.
Trippi Cue
An interview copilot that works inside the call: it hears the question addressed to you, suggests a short answer, and translates the conversation underneath. No bot joins, and it is a browser extension rather than an application.
Answers while the interview is still running.
Trippi Cue is in the Chrome Web Store and starting is free. One click on the icon when the call begins.
