Skip to content

Your Own AI v0.3.0

Choose a tag to compare

View all tags
@github-actions github-actions released this 13 Aug 02:06
· 53 commits to main since this release
v0.3.0
633a634

Muse Glimmer 30B, on your own hardware

  • Meta's Muse Glimmer 30B joins the offline catalog - chat, image understanding, and real agent work in your project folders, running entirely on your device
  • Works on NVIDIA and AMD consumer graphics cards (optional high-performance NVIDIA engine available in-app)
  • Inference engine updated to llama.cpp b10355
  • Reasoning strength adapts to the question - quick answers stay quick, hard problems get deep thinking

Bring your history

  • Import your conversations from ChatGPT, Claude, and Perplexity exports
  • Import coding sessions from Claude Code, OpenCode, Codex, Cursor, and Aider - most auto-detected with one click
  • Adopted conversations keep their original dates, and your AIs remember what's in them
  • Your Memory page: filter learned facts by import, forget any batch in one click

Online models

  • Grok 4.6 and Grok 4.6 (Web) available the day they launched - with support for the restored reasoning dial
  • DeepSeek V4 now serves its full 1M context; catalog shows every model's real context and release date
  • Online chats stream noticeably faster - connection reuse, lighter checks before the first token
  • Model cards show which shelf each model belongs to (chat, web search, coding) and sort correctly by newest

Smarter on your hardware

  • Context size is now chosen from your graphics card and memory together - agent sessions get the room they need (fixes project work dying at 8K context on 32GB machines)
  • Downloaded model cards show how each model fits this machine: full speed, runs slower, or too large - plus trained context vs what it runs at here
  • Auto model picks are fit-aware end to end: a model that struggles on your hardware hands off to one that runs at full speed, except while a project is open so your session's model stays warm

Diagnostics and reliability

  • One-click diagnostic report (Settings > Help & diagnostics) - system, models, routing decisions, crash records; redacted, saved locally, never uploaded; also copyable straight to the clipboard
  • If the app ever gets stuck starting, the loading screen offers to save the same report - no working app required
  • The report lists every model file's health, so a damaged download can't hide
  • One app instance, enforced - a second launch or a leftover process can no longer break startup
  • Faster startup on machines where the record-keeping engine takes time to warm up
  • The NVIDIA engine and the Your Own AI Build agent both gained proper update flows for future releases
  • Copy buttons work reliably on Windows everywhere in the app
  • Windows installer is code-signed; an MSI package is now published alongside it

Full download options and checksums below. Everything local stays free, forever.

Assets 9
Loading