Stop keeping four AI accounts.
One chat, a shelf of open models underneath it, and your history in one place instead of scattered across four. Ask for an answer, a working widget, or a designed deck — and every reply is labeled with the model that wrote it. $20 a month, flat.
- One workspace, many engines
- Every reply names its model
- Never trained on your chats
Design our Q3 investor update from these numbers — ten slides, editorial, charts where the story needs them.
Your deck is ready — ten slides, built from the numbers you attached.
Retention leads, revenue carries the middle, and the appendix holds the tables. Open it in the pane to reorder or restyle any slide.
- Llama 4 ScoutUnited States
- Gemma 3 27BUnited States
- GPT-OSS 120BUnited States
- Mistral Small 3Europe
- Qwen 3.5 35BChina
- Qwen 3.5 27BChina
- Qwen 3.5 397BChina
- Kimi K3China
- Kimi K3 FastChina
- GLM 5.3China
- DeepSeek V4 ProChina
- Qwen 3.6 27BChina
What you get back
Three kinds of output, from one conversation.
You do not pick a mode first. Ask for what you want and the answer arrives in the shape the question actually needs — a paragraph, something you can click, or a finished document.
Answers
A straight answer, and you know who wrote it
Streamed as it is written, with code highlighted and maths typeset. Every reply carries the name of the model that produced it and a badge for where that model was built — not 'our AI', the actual model, on every message.
How it worksLive widgets
Working things, not descriptions of them
Some questions want something you can poke at. Ask for a calculator, a chart you can toggle, or a timeline you can scroll, and it runs in the panel beside your chat — the working thing, not a screenshot and not a code listing.
How it worksDecks
A presentation the model designed
Ask for a deck and get one: the larger models lay out every slide themselves, and the free models write the words while Vresk's layouts do the design. Revise it by asking, step back through versions, and take it away as a real PowerPoint file with real text boxes.
How it worksYour history is never the hostage.
In Vresk, what you've written doesn't belong to any one model. Your conversations and what the product remembers about you live in your account, separate from whichever model happens to be answering. When a new open model joins the shelf and your plan includes it, you can start using it — and none of your own material has to move, because it was never inside the old one.
The shelf changes. Your setup doesn't.
Open models improve in different directions, and the interesting one is rarely the same two months running. Vresk keeps them on one shelf under one chat, so trying another one your plan includes costs you a click — not a new account, a re-upload, and a fresh set of preferences to teach.
How it works
One chat on top. Any engine underneath.
Start one conversation
Open Vresk and ask. You get one clean chat that streams answers as they're written and saves every conversation. Whatever engine is running underneath, this part never changes.
Pick the engine that runs
The picker is a curated shelf of open models, each with a plain-English note on what it's good at and a badge showing where it was built: US, EU, or CN. Pick the one that runs your chat; Vresk remembers your last pick in that browser.
Pick a different engine next time
Choose the engine for the job at hand — a demanding deck is not the same ask as a quick question. What Vresk remembers about you belongs to your account, not to a conversation, so it is there whichever engine you start with. Every reply carries an 'Answered by' label naming the model that wrote it.
Straight answers about your data
Know which model answered, where it was built, and what happens to your words.
Vresk doesn’t ask you to trust a black box. Every answer names the model that wrote it, every model wears its origin on its sleeve, and we make one hard promise about your words: they never train AI models.
- Every answer is labeled
- Each reply carries an 'Answered by' label naming the exact model that wrote it. Not 'our AI.' The model, by name, on every message. No mystery engine, no silent swaps.
- Never used to train AI models
- Your conversations and your questions are never used to train AI models. Not ours, not anyone else's.
- Origins on the label, filter in your hands
- Every model on the shelf carries a badge showing where it was built: US, EU, or CN. The default shelf is global. Prefer a tighter line? One switch limits the shelf to US- and EU-built models — and flipping it is entirely your call.
Your conversations are saved in your Vresk account; that’s what lets a new engine pick up an old thread. The no-training promise covers all of it: every conversation, on every engine. Models come and go. Your account stays.
Why open models
Open models are why there’s a shelf at all.
Open models publish their weights, so many companies can run them. That’s what lets Vresk put many on one shelf and charge one monthly price for the whole shelf.
No single company controls them
The weights are published, so no single company controls who runs them. No model on Vresk’s shelf is hosted by the company that made it.
New models join the shelf as they ship
The shelf is curated, not fixed: new open models are added as they ship, each with a plain-English note on what it is and a badge showing where it was built.
Your history is never locked inside one model
Your conversations and what Vresk remembers about you live in your account, not inside any one model. Pick a different model for your next task and none of it has to move.
Pricing
Every engine on the shelf. One flat price.
Try it free, or get Plus.
Plus
One number. No credits, no per-question fees.
- The full model shelf — pick the engine that suits each task
- Answers, live widgets and designed decks
- Your history and memory stay in your account
- Your chats never train AI models
Free to try, no card.
Get Plus — $20 a monthFAQ
Questions
Do I have to pick a model before I start?
Yes. The button beside Send reads “Pick a model to run” until you choose one your account can use, and Vresk remembers your last pick in that browser for your next chat. A designed deck asks far more of a model than a quick question does, so pick for the work in front of you. Every reply tells you which one answered.
Can I switch engines?
Yes, and the useful unit is the task rather than the sentence. Choose the engine for the piece of work in front of you; a new engine mid-thought spends its first move re-reading everything above it and buys you very little. What carries between tasks is the part that is yours — what Vresk remembers about you — because that lives in your account rather than inside any one model. If you want to compare, run the same ask across several engines at once and read them side by side, each pane labeled with the model that wrote it.
Why open models?
Open models publish their weights, so many companies can run them and no single company controls who gets to use them. That's what lets Vresk put many on one shelf, add new ones as they ship, and charge one flat price. Your history lives in your Vresk account, not inside any one model, so when a better open model arrives and your plan includes it, you just pick it. Every reply names the exact model that wrote it, and your conversations are never used to train AI models.
What can it actually make?
Three things, and you do not choose a mode first. Answers in the thread, with code highlighted and maths typeset. Live widgets — a calculator, a chart you can toggle, a timeline — that run in the panel beside your chat. And decks — on the larger models the model designs each slide itself, on the free models Vresk's slide layouts do the design — which you take away as a real PowerPoint file. A written document downloads as a PDF or as Markdown.
Which models are on the shelf?
Today's shelf is DeepSeek V4 Pro, Gemma 3 27B, GLM 5.3, GPT-OSS 120B, Kimi K3, Kimi K3 Fast, Llama 4 Scout, Mistral Small 3, Qwen 3.5 27B, Qwen 3.5 35B, Qwen 3.5 397B and Qwen 3.6 27B. Each one carries a plain-English note on what it is and a badge showing where it was built. The shelf changes as new open models ship — that's the point.
Is it really one flat price?
Yes. Plus is $20 a month, with no credits to top up and no fee per question. It includes a weekly allowance, and your account shows how much is left and when it resets.
Do you train AI on my conversations?
No. Your conversations and questions are never used to train AI models — ours or anyone else's. That holds on every engine on the shelf.
Some of these models were built in China. What does that mean for me?
The origin badge tells you where a model was built, so you can decide what you're comfortable with. The default shelf is global. If you'd rather skip CN-built engines entirely, one switch limits the shelf to US- and EU-built models, and flipping it is entirely your choice. Wherever a model was built, your conversations never train it.
Does it remember me between conversations?
Yes, when memory is on. You can turn it off entirely, in which case Vresk saves nothing about you between conversations and uses nothing it saved before, and one button in Your data makes it forget everything it saved. You can also choose how long conversations and files are kept, and delete them.
What happens if something takes a long time to build?
A single turn has a time limit, and an ambitious deck or build can exceed it. When that happens Vresk keeps working in the background on its own and tells you so under the reply, with a Stop button if you'd rather it didn't — you can close the tab, and the finished result lands in the conversation. If an answer stops partway through, what it already wrote is kept rather than discarded, so you can continue from there.
How do I pay?
Sign in to an account that has confirmed it is 18 or older, then open Account → Plan and choose Get Plus. You pay by card on Stripe's page, so your card details go to Stripe, not to Vresk. Plus is $20 a month and renews automatically each month until you cancel. Cancel any time from your account page. You keep Plus until the end of the month you paid for. Plus is sold to billing addresses in the United States and Canada. If you subscribe with a billing address anywhere else, we cancel and refund the payment automatically. Plus isn't available in Quebec yet. If you subscribe with a Quebec billing address, we cancel and refund the payment automatically.