Skip to content

How Kyra responds

Kyra is short by default. Most replies are a sentence or two: the outcome first, then silence. That is a deliberate design choice and occasionally an annoying one, so it is worth knowing how to override it.

Headline first. The answer, then the reasoning — never the other way round.

Long things are not read aloud. Ask for a draft, a list, or a detailed explanation and you get a short spoken summary plus the full thing on screen. Listening to a document read out is nobody’s idea of a good time.

Silence is never left unexplained. If something will take a while, Kyra says so and hands control back rather than going quiet. Background work →

Outcomes are always stated. Saved, failed, queued, waiting for approval, or needs setting up — you get told which, briefly. “Saved locally, but the sync failed — it is safe to retry” rather than a cheerful “Done!”.

These work in any conversation, at any point, including over the top of Kyra talking.

SayEffect
“Short version” / “TLDR”Collapse to one sentence.
“Go deeper” / “Explain more”Expand the current point one layer.
“Skip ahead” / “Next”Move past this section.
“Draft it” / “Put it in writing”Stop discussing, start producing.
“Save that” / “Put that in the doc”Persist what was just said.
“Cancel that”Cancel the most recent task.
“What are you working on?”List everything in flight.
“What needs approval?”Just the things waiting on you.
“What do you remember?”Summarise stored memory, with a way to delete.
“Forget that”Delete a memory — Kyra confirms which one first.

Kyra shifts how it talks based on what you are doing. You do not have to manage this, but naming the register you want is a fast way to change the whole shape of a conversation.

RegisterIt sounds likeTrigger it with
QuickOne or two sentences, doneThe default for short requests
ExplainingLayered — intuition, then mechanics, then example“Explain how X works”
Designing with youDrafts first, states assumptions, asks only blocking questions“Design this with me”
DraftingProduces the artefact, does not read it out“Write me a…”
ReviewingDecision first, then trade-offs“Compare these”
ApprovingOne clear question, one consequence, yes or noAutomatic for anything consequential
DrivingStrictest — voice only, no detailAutomatic on a car call

For harder questions Kyra can spend longer reasoning before answering. It is slower and usually better, and you can control how much of it happens.

Portal → Settings → Thinking sets your default. Kyra will also tell you when it is thinking rather than leaving you with dead air.

The trade-off is real in both directions: more thinking means better answers on hard problems and needless delay on easy ones. The default is tuned for everyday use — turn it up when you are working on something that deserves it.

Ask how Kyra is built and you get a genuine but high-level answer. It will not name the services, models, or vendors behind any part of it, and pressing does not change that.

This is not evasiveness about what it does with your data — that is documented in your data and answered plainly. It is a line about implementation detail.

If you are building something similar, Kyra will happily help with the engineering question in general terms.