Getting started

1. 1

   Open the chat

   Go to the chat. There is no setup step, no model to configure first, and nothing to install. A new conversation starts empty with the composer focused. The first time you use Vresk, a one-line bar asks you to confirm you’re 18 or older (on a phone, the question appears the first time you send, attach or dictate) — it is a one-time self-confirmation, and anything you try to send first simply waits until you tap I’m 18 or older.
2. 2

   Ask for something you actually want

   Not a test question. The product is easiest to judge on real work, and the interesting behaviour — building a widget, designing a deck — only shows up when the ask needs it.
3. 3

   Watch which engine took it

   Every reply carries an Answered by label naming the model that wrote it, with a badge for where that model was built (on a phone, the model’s name alone, beside a small square). Nothing is swapped underneath you without the label changing.
4. 4

   Keep going in the same thread

   Follow-ups read everything above them. If the answer produced something in the side panel, asking for a change edits that thing rather than starting a new one.

Three shapes, and Vresk picks between them from the ask itself — there is no mode to select first:

• An answer — prose in the thread, streamed as it is written. This is most of them. More on answers.
• A live widget — a working, interactive thing that runs in the panel beside the chat: a calculator, a chart, a timeline. More on widgets.
• A deck or document — slides designed for you (the larger models lay each one out themselves; on the free models you add Design under the + button, see the plan first, and the deck is text only for now), or a written document, exportable as a file. More on decks.

Three more shapes feed these when the ask calls for them: a quick web search, a deep research run that comes back as a sourced report, and images, generated or edited in place.

Six things are worth knowing on sight.

• The engine picker sits at the top of the conversation. It shows the shelf, what each model is good at in plain language, and where it was built.
• The attribution chip under each reply names the model that wrote it. On a side-by-side comparison, each pane carries its own.
• The side panel opens by itself when a reply produces something that is not prose. It holds the widget or deck, its versions, and its copy and export buttons.
• The thinking clock appears while a model is working before it starts writing. Some models think for a while first; that is the model, not a stall.
• The intent menu is the sparkle button in the composer, reading Add. Vresk reads your message and picks how to answer: ask for a web look-up, proper research or a calculation in your own words and it does that. The one path you add by hand is a designed deck or document — Design, in that menu (on a phone, under the + button). Adding it pins a dismissible chip for your next message; tap it to go back to automatic.
• The capability chips on an empty conversation show what Vresk can do. Tapping one arms the composer for that kind of ask and puts the cursor in the input — it never writes your message for you.

Conversations are saved to your account and listed in the sidebar. So are the artifacts they produced, with their version history — reopening a conversation reopens the deck you were editing, at the version you left it.

Separately from any one conversation, Vresk keeps what it remembers about how you work. That layer is the part that does not care which engine is serving you, which is the whole reason it is kept apart from the thread. More on memory.