Vresk is an AI work platform that puts multiple open-source models under the same chat interface. Users only need to start a conversation, and they can get three forms of output in the same window: direct answers, interactive widgets, and designed presentations.
Main Capabilities
One conversation, three outputs Users do not need to choose a mode first. They simply make a request, and the answer appears in the form the question itself requires—a piece of text, something clickable and operable, or a finished document.
Direct answers, with the author identified Answers stream out as they are written, code is highlighted, and mathematical formulas are typeset. Each reply is labeled with the name of the model that generated it, as well as a badge showing where that model was built, rather than being collectively called "our AI."
Interactive results, not descriptions Some questions require something that can be operated by hand. Users can ask Vresk to generate calculators, toggleable charts, or scrollable timelines, and they actually run in the panel next to the chat, rather than being screenshots or code listings.
Model-designed presentations Users can request a presentation. For larger models, the model designs each slide itself; for free models, Vresk's layout handles the design. Users can revise it through conversation, step back through versions, and export it as a PowerPoint file with real text boxes.
History is not locked into a single model The user's conversations and the product's memory of the user are stored in the account, independent of the model that answered at the time. When new open-source models are added and included in the plan, users can start using them directly, and their materials do not need to be migrated, because they were never inside the old model.
The model shelf changes, the setup does not Open-source models advance in different directions, and interesting models rarely stay the same for two consecutive months. Vresk puts them on the same shelf under the same chat. Trying another model in the plan takes one click, rather than re-registering an account, re-uploading, and resetting preferences.
How to Use
- Start a conversation: open Vresk and ask directly to get a clean chat interface, with answers streaming out, and all conversations saved.
- Choose the engine to run: the model selector is a curated shelf of open-source models, each with a plain-English description of what it is good at, as well as a badge indicating where it was built (the United States, the European Union, or China). After choosing the model to run the current chat, Vresk remembers the last choice in that browser.
- Choose a different engine next time: choose the engine according to the task at hand—if you wantA tall presentation and a quick question are not the same thing. Vresk's memory of the user belongs to the account, not to a single conversation, so it is there no matter which engine you start with.
Pricing and subscriptions
Vresk offers a free trial with no card required. The Plus plan is $20 per month, with no credits to top up and no per-question charges. Plus includes a weekly quota, and the account page shows how much is left and when it resets. Subscriptions require signing in to an account confirmed to be at least 18 years old, choosing Get Plus in Account → Plan, and paying by card through the Stripe page. Plus renews automatically each month until the user cancels; it can be canceled at any time on the account page, and the remainder of the paid month remains usable. Plus is currently only sold for billing addresses in the United States and Canada. Subscriptions with billing addresses in other regions are automatically canceled and refunded. Quebec is temporarily unavailable.
Privacy and data
Vresk states: the user's conversations and questions are never used to train AI models, whether its own or anyone else's, and this commitment applies to every engine on the shelf. Source badges tell the user where a model was built, and the default shelf is global; if the user wants to skip China-built engines entirely, there is a toggle that can restrict the shelf to models built in the United States and the European Union. Memory can be turned off completely, and once it is off Vresk will not save any information about the user between conversations or use anything previously saved; in "Your data" there is a button that lets it forget everything saved. Users can also choose how long conversations and files are retained and delete them.
Current model shelf
Today's shelf includes 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 model comes with a plain-English description and a badge marking where it was built. The shelf changes as new open-source models are released.
Limitations and caveats
A single turn has a time limit, and a demanding presentation or build may time out. When that happens, Vresk keeps working in the background and says so below the reply, and the user can choose to stop it; even if the tab is closed, the completed result will appear in the conversation. If an answer stops midway, the part already written is kept rather than discarded, and you cancontinue from there. As of the time the information was provided, details such as the specific quota for the free tier, availability on other platforms, and the specific languages supported have not yet been provided.
