One copilot, six ways to help, chosen by what you are doing on the call. Meeting mode is the default. The others are a tick box in the Chrome extension or the Windows app, and every one of them uses the same account, the same minutes and the same history.
Meeting mode
The default. Loquelle hears the other side through the tab or your device audio, spots a question aimed at you the moment it is asked, and hands you an answer you can say out loud in your own voice, grounded in your brief and the document you attached. It stays silent while the other side is mid-explanation: a card every few minutes is right, a card every poll is noise.
Grounding: a number, a name or a subject that appears nowhere in your document, your brief or the transcript never reaches a card. Ask box: type a note mid-call ("they misunderstood pricing") and the next card answers it first. Continue: pick a past meeting and today's call starts with last time's recap as background and the open follow-ups from that thread.
Technical mode
For coding and design interviews. When the interviewer poses a task, Loquelle writes one clean implementation in the language being used, with a spoken lead-in you can say to introduce it and a comment on every non-trivial line giving the reason, so "why did you do that" has an answer. Design questions come back as components, data model, flow and trade-offs; diagram requests as ASCII or Mermaid.
Incremental: when the interviewer adds a method or changes a constraint, you get only the new piece and where it goes, never the whole solution again. Runs on the larger model for defensible code. Works in civil, electrical and mechanical engineering, finance and other non-software fields too.
Presentation mode
Tick it, attach your deck (.pptx with speaker notes, or PDF), set the time slot, press Start. A strip shows the slide Loquelle believes you are on and a clock against your slot. Each new slide gets a spoken script long enough to fill its share of the slot; say "next slide" in any language and it moves on. Audience questions are answered from the deck, in the language they were asked in, with the slide number.
Cues: halfway, ninety percent, time up, two slides behind pace, and one line to close the discussion well. Prep me first and you get a full script for every slide before you start.
Screen mode
Press Ctrl+Shift+A in the Windows app, or Answer this tab in Chrome, and Loquelle reads what is visible and answers it: a written question, a multiple-choice test one item at a time, a coding task with complete runnable code, an error message with the likely fix. Works mid-meeting (the answer joins your card stream and is cross-matched with what the call is discussing) and on its own outside meetings.
Privacy: the capture is processed in memory and never stored; the answer appears only in your private overlay, invisible on screen share. Every failed attempt is refunded.
Listening mode
Tick "I'm only listening" for a webinar, broadcast or recording. Loquelle switches from words-to-say to notes: the specific claims, figures, product names and offers just stated, quoted faithfully so you can verify them later, and now and then a sharp question to ask the speakers afterwards. It also turns itself on when your brief says webinar, lecture or podcast, or when a dozen polls pass with nobody addressing you.
At Stop: key takeaways, every claim and number, questions to ask, a note on why it matters to you, emailed as "Your notes". No email draft, because there is nobody to send one to.
Learn mode
Tick "I'm in a lesson" and type the course name. In a lecture, a course video or a training call, Loquelle takes study notes as it plays: Definition when a term is defined, Why it matters, Example, On the test, Ask. At Stop you get concepts in the order they were covered, definitions in the teacher's wording, examples, what was flagged as examinable, six to ten flashcards and a five-question quiz, built only from the transcript and emailed to you.
Course threading: each lesson is saved under the previous one of the same course, so week four is read with weeks one to three as background. Guards: no words to say, no cue logic, and a flashcard whose answer was never said is dropped.