Answers

Most of what Vresk produces is an answer in the thread. It streams as it is written, it is labelled with the model that wrote it, and it stays in the conversation for the next question to build on.

The 'Answered by' label

Every reply carries a chip naming the model that produced it and a badge showing where that model was built. This is not decoration and it is not a category — it is the specific model, by name, on every message. On a phone the label is shorter: the model’s name, beside a small square under the reply.

It exists because the alternative is the norm elsewhere: a product says “our AI” and quietly serves you whatever is cheapest that minute. If a different engine answers, the label changes. Open the label to see why that model answered — the popover and picking a model are covered in Choosing an engine.

On a comparison

When you run the same question across several engines, each pane keeps its own label. The panes are separate answers, not one answer merged.

Streaming, thinking, and stopping

Answers arrive as they are written. Before the first word there may be a pause with a clock on it — that is the model thinking, and on the heavier reasoning models it can be a genuinely long pause. It is not a stall, and Vresk no longer treats it as one.

What you see	What it means
A clock, no text yet	The model is reasoning before it writes. Normal on the deeper engines.
Text appearing steadily	The answer is streaming. You can read along; nothing is buffered until the end.
An answer that stops mid-sentence	The engine stopped early. What it already wrote is kept, not discarded — you can continue from there or ask again.
A notice that the work ran long	The ask needed more time than a single turn allows. See the note below.

Long work does not have to be lost

When something ambitious runs past the time a single turn gets, Vresk keeps working on it in the background and says so under the reply. You can leave the page; the result lands in the conversation. More in troubleshooting.

Steering a reply

When your message could have been read as wanting more than an answer, the reply carries a row of one-tap chips — “Look it up”, “Research it properly” — each with an honest note of what it costs in time. A chip re-runs the same message the other way; you do not retype anything. The chips are saved with the reply, so you can come back tomorrow and still take the offer.

After some answers you will also see follow-up chips — “Shorter” when an answer ran long, “Look it up” when an answer admits it may be out of date. These send a short follow-up on your behalf; hover one to read the exact sentence it will send before you tap it.

While an answer is streaming, the Send button becomes Stop. Stopping keeps everything already written — the reply is marked as stopped and saved as far as it got, and you can continue from there.

Following up well

A follow-up reads the whole thread above it, so you rarely need to restate context. Two habits make a large difference:

Ask for the change, not the whole thing again. “Make the third section shorter” is faster and more accurate than re-describing the document. When something exists in the side panel, a narrow ask is edited in place rather than rebuilt from scratch.
Say what is wrong, not just what you want. “This assumes I already have the data” gets you further than “try again”, because the second one leaves the model guessing which part missed.

Editing your message

Any message you sent has an edit control (hover or focus the message to reveal it; on a phone, tap the message and Edit appears under it), including messages in the middle of a thread. Editing rewrites the question and re-runs it from that point.

Because the answers that followed were replies to a question that no longer exists, they are removed with the edit — and the editor says so out loud before you save, with the exact count of messages that will go. The number shown is the number removed, or nothing is.

The editor stays open until the rewrite has actually landed, so a refusal can never lose your words — they stay in the editor, exactly as you typed them. An edit you did not change anything in simply closes; it never re-runs a good answer for no reason.

Code, maths and tables

Answers render as rich text, not a plain wall. Code blocks are syntax-highlighted and carry a copy button. Mathematical notation is typeset. Tables and lists render as tables and lists.

Longer code — something you would actually save to a file — usually moves out of the thread and into the side panel, where it gets its own copy and download controls and its own version history. More on artifacts.