Introducing the new VexAI: an assistant with its own computer, rebuilt from the ground up

The new VexAI gives every chat a computer that makes real files, runs Python with the libraries you know, draws live charts and diagrams, keeps working when you switch chats, and works with your own models and tools. Rebuilt on the new VexAPI and Vex ID, in the new Arioron style.

14 October 2026 · 4 min read

VexAI is Arioron Labs' AI assistant. This release is a complete rebuild. VexAI now runs on the new VexAPI and signs you in with the new Vex ID, and almost every part of it is new: how it answers, what it can do on its own, and how it looks. We set out to build the assistant we wanted to use every day, one that doesn't just describe the work but actually does it.

Here is what's new.

A computer for every chat

Ask for something you can download and VexAI turns on a computer for the chat: a private, sandboxed workspace where it writes and runs code on real files.

Real files, not descriptions. Ask for a zip of a project, a LaTeX paper, a Word report, a CSV, a cleaned spreadsheet or a converted file, and it appears under the answer as a download.
Bring your own files. Upload archives, code or data of any type and VexAI unpacks, edits, renames, organises or analyses them.
Watch it work. Every step the computer takes appears as a live code cell, showing the code as it is written and the output as it runs. Copy any of it with one click.
Install packages. VexAI installs Python packages from PyPI with their dependencies (up to 200 MB per computer), including ones like requests, rich, beautifulsoup4, openpyxl, sympy and reportlab.
It waits for you. A computer that isn't being used is stopped after a few minutes with its files saved. Come back to the chat and it starts again with everything where you left it, and earlier download links keep working.

Data science that runs in your browser

numpy, pandas, matplotlib, seaborn, scipy, scikit-learn, statsmodels and more now work in every chat.

The libraries you already know. Ask VexAI to analyse a spreadsheet, fit a model or plot a trend and it writes ordinary pandas and matplotlib code.
Runs on your machine. That code runs in a secure sandbox in your own browser, using WebAssembly. Your chat's files and attachments are available to it, and the results are saved back to the chat's computer.
Charts appear in the answer. Every figure is saved as an image and shown under the reply, ready to download.
More on demand. Other Python packages from PyPI install automatically when the code needs them.

The first run in a session takes a few seconds while the libraries load. After that, runs start almost instantly.

Answers you can see

When a picture says it better, VexAI shows rather than tells. It writes interactive visuals straight into the answer, styled to match the app.

Charts: bar, line, area, pie, donut, scatter and radar, with stacked series, dual axes and target lines, plus graphs of mathematical functions.
Key numbers as stat cards with change indicators and sparklines.
Timelines, side-by-side comparisons with pros, cons and scores, callouts, and quizzes you can answer right in the chat.
Diagrams for processes, architectures, sequences, state machines, data models, mind maps and project plans, themed for light and dark mode.
Code that reads properly, highlighted with the same grammars as VS Code in hundreds of languages, plus typeset math, tables and clean Markdown throughout.

VexAI only charts real numbers: yours, ones its tools found, or well-established facts. It never invents data to fill a chart.

Canvas: build and write side by side

Canvas opens a workspace next to the chat for longer pieces of work, such as a document, a block of code, a web page, an SVG or a diagram.

Live previews. Web pages and apps run as you watch, so you can ask for "a pomodoro timer" and use it a moment later.
Edit it yourself. Change anything by hand and VexAI picks up your edits in its next reply.
Every version kept. Step back through earlier versions at any time.

Research, browsing and learning

Web search with sources. Answers about anything recent come with the links they used.
Deep research. VexAI searches widely, reads the sources and writes a long, cited report on the question you give it.
Browse. VexAI can drive a real web browser, opening pages, clicking, typing and scrolling, and shows you each step as it goes. It never signs in, buys anything or submits a form unless you asked for exactly that.
Learning mode. VexAI becomes a patient tutor that explains step by step, checks your understanding and helps you work things out rather than handing you the answer.
Think longer. For hard problems, give the model more time to reason before it answers.

VexAI decides for itself when to search, use the computer or open the browser. You don't have to switch anything on.

It keeps working when you leave

Start a long task, switch to another chat or close the tab, and the reply keeps going. Chats that are still working show it in the sidebar. Open one again and you pick up the reply live, exactly where it is.

Talk to it

Voice mode for a spoken back-and-forth conversation.
Dictation to speak instead of type.
Read aloud on any answer.

It remembers what matters

Memory. VexAI remembers useful things you tell it, such as your preferences, your projects and how you like answers. You can see, add, pin or delete every memory in Settings, or turn memory off entirely.
Personalisation. Tell VexAI how to address you and how to respond, and it follows that in every chat.
Projects. Group related chats, give them shared instructions, and add knowledge files that VexAI searches when they're relevant.

Bring your chats with you

Coming from ChatGPT, Claude or Grok? Export your data there, upload the file to VexAI, and your conversations come across as VexAI chats, ready to continue. VexAI can also carry over what those assistants had learned about you, so your memory doesn't start from zero.

Your tools and your models

Connectors. Add any remote MCP server (the open Model Context Protocol) and VexAI can use its tools in chat, alongside its own.
Your own models. Use models from Anthropic (Claude), OpenAI, Google Gemini, OpenRouter, Groq, Mistral, DeepSeek, xAI (Grok) or any OpenAI-compatible endpoint with your own API key. They appear in the model picker next to Arioron's models and get the same tools: web search, the computer, the browser, deep research and your connectors. Your keys stay on our servers and never go back to the browser.
Arioron models. The picker lists every model available on VexAPI, including Perceptix Vex Amber and Vex Photon Lite, so new releases appear as soon as they ship.

Every chat, your way

Edit any message, regenerate an answer, or regenerate with another model, and flip between the versions.
Branch any answer into a new chat to explore a different direction.
Suggested follow-ups after each answer.
Share a read-only link to a chat.
Temporary chats that are never saved, and a one-click export of all your data.
Search every chat with Ctrl + K, and start a new one with Ctrl + Shift + O.

Sign in with your Vex ID, the one account for everything Arioron, or with Google or email. Your existing VexAI chats and settings carry over to the new app.

A new look

VexAI now matches the rest of Arioron: ink on paper, soft edges, calm motion and nothing that shouts. The light and dark themes were designed side by side, the layout adapts from wide monitors down to phones, and the whole app shares one set of design tokens with arioron.com, Vex ID and VexAPI.

Get started

Go to vexai.arioron.com and sign in with your Vex ID.
Ask for something real: "analyse this spreadsheet and chart the trend", "turn these notes into a Word report", or "build me a habit tracker".
Coming from another assistant? Open Settings → Import and bring your chats and memory with you.
Add your own models and connectors in Settings → Models and Settings → Connectors.

We can't wait to see what you do with it.

The Arioron team

Five companies, one engineering standard.