Introducing the new VexAPI

VexAPI is Arioron's AI platform for developers. One API key gives you our models, agents, voice, retrieval and sandboxed compute. This release is the biggest update since launch. It adds a full app studio called Build, a library of natural-sounding voices, and realtime voice conversations. It also gives VexAPI a whole new look: the console has been redesigned from top to bottom to match Arioron's new visual identity.

Here is what's new.

Build: describe an app, watch it get written, run it live

Build turns a prompt into a working web app inside your browser.

Describe what you want ("a habit tracker with streaks and a 12-week heatmap", "a recipe app that suggests meals from what's in my fridge"). Arioron writes a complete React + TypeScript + Tailwind project in front of you. Files appear in the editor as they are written, and the app starts running in a live preview the moment the code is ready.

  • A real editor. You get a file tree, a full code editor with syntax highlighting, and a Save & run button. Edit any file yourself and the preview restarts instantly.
  • Keep refining in chat. Ask for changes in plain language ("make the cards draggable", "add dark mode") and Build updates only what needs to change.
  • Show, don't tell. Attach screenshots or mockups (paste or drop them in) and Build uses them as a reference.
  • One-click fixes. If the app throws an error, it appears in the built-in console with a Fix button that sends the error straight back to the builder.
  • Suggested next steps. After each build, Build proposes a few sensible improvements you can apply with a click.
  • Checkpoints. Every build and save is kept. You can restore any earlier version, and restoring never deletes anything.
  • Device previews. Check your app at desktop, tablet and phone widths, or go fullscreen.
  • AI inside your apps. Apps you build can call Arioron's models themselves: chat, summarisation, image understanding and more. The platform handles this securely, so no keys ever end up in the app's code.
  • Share with a link. Publish an app and anyone can use it at its own URL. Visitors can remix it into their own copy. AI usage from shared apps is billed to your key and rate-limited per visitor, so a popular app can't run away with your budget.
  • Download the code. Export the whole project as a ZIP and keep building anywhere.
  • Pick your model. Choose which Arioron model writes your app, and separately which model powers the app's own AI features.

Build runs every app in an isolated sandbox. Nothing an app does can touch your account or the console.

A new look: the console, rebuilt in the Arioron style

VexAPI now looks and feels like the rest of Arioron. We rebuilt the console from the ground up on Arioron's new design language: ink on paper, soft edges, and nothing that shouts. Your tools should feel calm enough to work in all day and polished enough to be proud of.

Ink and paper

The new Arioron identity is mostly monochrome: deep ink on clean white, with colour saved for what matters. The console follows the same rule.

  • Ink-black primary actions and a quiet greyscale interface mean your eye always lands on the one thing you can do next.
  • Colour carries meaning. Green means success, amber needs attention, red is an error, and a small coloured dot tells model families apart at a glance. Nothing is coloured just for decoration.
  • The Arioron wordmark sits at the top of the sidebar, and the sign-in and landing pages share the faint grid texture used across Arioron's site.

Soft, rounded and layered

  • The console sits on an inset canvas: a soft grey frame holds the sidebar, and your work lives on a raised white panel. You always know where the navigation ends and your content begins.
  • Rounded corners everywhere, from buttons and inputs to cards, dialogs and the Build editor, using one consistent set of radii instead of a mix of sharp and soft.
  • Gentle depth. Feather-light shadows lift cards and menus just enough to show what sits on top. There are no heavy borders or harsh outlines.
  • Calm motion. Menus, popovers and dialogs ease into place in a fraction of a second, a slim progress bar shows when a page is loading, and the "thinking" indicator breathes softly while a model works. All of it respects your system's reduced-motion setting.

Type that reads well

  • Inter for everything you read in the interface, set with tight, modern headings and comfortable line spacing for long answers.
  • JetBrains Mono only for code, IDs and keys, so identifiers stand out from prose.
  • Model answers are rendered as proper documents: real headings, lists, tables, typeset math and highlighted code, all in the same visual style as the rest of the console.

Light and dark, done properly

Both themes were designed side by side, not generated from each other. Dark mode uses layered deep greys rather than flat pure black, so cards and menus still read as separate surfaces. Code blocks, charts, badges and the Build editor all switch cleanly. VexAPI follows your system setting, and you can switch any time from the top bar.

Easier to find your way

  • A reorganised sidebar grouped by what you're doing: Library (models, voices, files, vector stores, memory), Agents, Tools, Pipelines, Monitor and Workspace.
  • Search or jump to anything with Ctrl + K.
  • Keyboard-first throughout: Enter to send, Ctrl + Enter to run, Ctrl + S to save in Build, Esc to close any menu.
  • Clear, honest states. Every list has a helpful empty state that tells you what to do first. Destructive actions such as delete stay quiet until you need them, and always ask before anything is lost.
  • Works on any screen. The console adapts from wide monitors down to tablets and phones.

One system, every page

Everything is built on a single set of design tokens for colour, type, spacing, radius, shadow and motion. The Playground, Build, Agents, the Voices catalogue, Containers, the docs and the admin pages all share one look, and every future feature arrives already matching it.

Pick the right model, everywhere

A new model picker appears across the console: in Build, the Playground, Conversations, Agents and Evaluations. Each model shows a short description and what it can do, such as vision, reasoning, tool use and context size, so you can choose with confidence.

  • Switch an agent's model straight from its page. It saves instantly.
  • The Playground and Conversations remember your last choice.

Our current line-up:

Model Best for
Perceptix Vex Amber Our flagship multimodal model for everyday production work
Perceptix Vex Amber 2.0 The next generation of Amber, in preview
Vex Photon Lite Fast, low-cost text at high volume

Rolling -latest aliases are also available if you always want the newest release.

Answers that read beautifully

Every place a model answers you now renders rich, sanitised output as it streams. That covers the Playground, Conversations, Agents and Build chat.

  • Markdown: headings, lists, quotes, links and tables.
  • LaTeX math: inline and display equations, typeset properly.
  • Code blocks: syntax highlighting with one-click Copy.
  • Thinking indicators: you can see when the model is thinking or using a tool, such as searching the web or checking the weather, before the answer starts to arrive.

A voice library with natural, open voices

The new Voices page is a catalogue of 38 voices across 7 languages and accents: American and British English, Spanish, French, Hindi, Italian and Brazilian Portuguese. Each voice has a short description of how it sounds ("warm and expressive", "deep and relaxed", "clear and professional") and a preview button, so you can hear it before you use it.

  • Text-to-speech via /v1/audio/speech. Pick any library voice by name; common voice names from other platforms are mapped automatically, so existing code keeps working.
  • Speech-to-text and translation via /v1/audio/transcriptions and /v1/audio/translations.
  • Correct pronunciation of names. Voices get brand and product names right, starting with our own: Arioron is said "A-rye-orun", the way we say it.

Realtime voice conversations

/v1/realtime lets users talk to Arioron and hear it answer back. Use it from your own app over a WebSocket, or try it in the console's Realtime page.

  • Speak naturally. The model answers when you pause, and you can interrupt it mid-sentence just by talking.
  • Replies start speaking as soon as the first phrase is ready, without waiting for the whole answer.
  • It uses any voice from the library, and you can type as well as talk.

Agents that use tools and remember

Agents combine a model, your instructions and a set of tools, then run multi-step tasks on their own.

  • Built-in tools include web search, page scraping and article extraction, weather, time, unit conversion, calculation, and file creation and reading.
  • Optional persistent memory lets an agent recall what it learned in earlier runs.
  • Control maximum steps and timeouts per agent, cancel runs in flight, and inspect every step of every run in the console.
  • Run agents with Ctrl + Enter and watch the result stream in, fully formatted.

The same capabilities are available through the API: /v1/agents and agent runs, /v1/responses with tool calling, and /v1/memories.

Retrieval with vector stores

New vector stores give your apps and agents long-term knowledge:

  • Upload files, group them into stores and search them semantically with /v1/vector_stores/{id}/search.
  • Add files one at a time or in batches.
  • Manage everything from the new Vector stores pages in the console, including uploads, file status and test searches.
  • Generate your own embeddings with /v1/embeddings.

Containers: sandboxed compute for code

Containers give your agents and applications an isolated workspace to run code and work with files.

  • Create a container, upload files, execute code and download the results.
  • Edit files in place and browse them from the new Containers console.
  • Everything runs in a sandbox, separate from the platform itself.

Built for production

The new VexAPI keeps everything you need to run AI in production in one place:

  • Familiar APIs. Drop-in compatible /v1/chat/completions, /v1/responses and /v1/messages endpoints (with token counting), plus /v1/models with full capability metadata. Bring your existing SDKs and change a base URL.
  • Streaming everywhere, with structured outputs and tool calling.
  • Conversations API to store and continue multi-turn threads server-side.
  • Batches and background jobs for large offline workloads.
  • Evaluations. Define test suites and run them against any model to compare quality before you switch.
  • Model routing. Create aliases that fail over automatically across models based on their recent error rates.
  • Webhooks that notify your servers when responses, batches and agent runs finish.
  • Browser and computer sessions for agents that need to work with real web pages.
  • Moderation via /v1/moderations.
  • Observability. Request logs, per-request detail, and usage by day, by tokens and by request, in the console and through /v1/usage and /v1/logs.
  • Projects and API keys to separate environments and teams.

Get started

  1. Sign in to the VexAPI console and create an API key.
  2. Open Build, describe an app, and press Build. Or try I'm feeling lucky.
  3. Explore the Playground, browse the Voices library, or create your first Agent.
  4. Read the Docs for endpoint references and code samples in your language.

We can't wait to see what you build.

The Arioron team