Skip to content

Changelog

What’s new in X Alpha AI

Every release of the desktop app, newest first.

  1. v0.1.0

    First installers

    • macOS: a .dmg for Apple Silicon. Test builds are ad-hoc signed. Open once with Open Anyway.
    • Windows: a setup .exe, built automatically for every change.
    • Faster image reading: images use about half as many tokens; a photo is read in about 2 seconds on a Mac.

  2. Files, images and every model on Hugging Face

    • Attachments: PDFs, Word, PowerPoint, spreadsheets, text, code and images, from the + menu, by dropping, or by pasting a screenshot.
    • The + menu holds Attach files, Web, Think and Browser; switched-on tools show as chips.
    • Models from Hugging Face: search all GGUF chat models; the version that fits your computer is picked for you.
    • Model settings: context window (automatic or your choice) and temperature.
    • New look: ERBrains brand colours and a cleaner chat layout.

  3. Browser control and built-in search

    • Browser control: a browser panel beside the chat. The AI reads, clicks, types and scrolls after you allow each site.
    • macOS support for the browser panel.
    • Built-in web search: works out of the box in a hidden browser window; your own SearXNG server is optional.
    • Faster web answers: the model reads less and writes less, measured on a test set of questions.

  4. Web search in the app

    • Web search: Web or /web searches and answers with clickable sources.
    • Links in your message are read for you.
    • /pkg gives exact latest package versions from the registries.
    • Allow / Deny before the AI searches on its own.
    • Automatic context window sized to your computer's memory.
    • General-purpose models in the list, each saying what it's for.

  5. The first X Alpha AI window

    • Chat with streaming answers, Markdown and code blocks with Copy, and Stop.
    • Saved chats that reopen after a restart; rename and delete from the sidebar.
    • Local models: the app checks your hardware, shows which models fit, downloads (with pause and resume) and runs them offline.