Skip to content

v1.0.5

Choose a tag to compare

View all tags
@avikeid2007 avikeid2007 released this 12 Jan 17:53
· 90 commits to main since this release
1.0.5
2c40ee6
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Release Notes - KaiROS AI (v1.0.5)

🚀 Highlights

📦 Massive Model Catalog Expansion

The model catalog has been expanded to 31 pre-configured models across 9 major organizations.

  • New Organizations: Qwen, Google (Gemma 3), Meta (LLaMA 3.1/3.2), Microsoft (BitNet, Phi-3), MistralAI, OpenAI (GPT-oss), RWKV, Stability AI, and SmolLM2.
  • UI Redesign: Features collapsible organization sections and a dedicated "Downloaded Models" section for quick access.
  • Advanced Filtering: New filters for Organization, Model Family (e.g., Qwen, Gemma), and Hardware Variant (CPU-Only, GPU-Recommended).

⚡ Hardware & Performance

  • Next-Gen GPU Support: Fixed VRAM detection for modern GPUs (RTX 50/40/30 series and AMD RX 7000/6000). Correctly identifies 16GB VRAM on systems like the RTX 5070 Ti.
  • LLamaSharp 0.25.0 Upgrade: Updated the core runtime to support the latest GGUF formats, including Qwen 3 and Gemma 3 models.
  • Smart Fallback Loading: If a model fails to load into GPU memory, the app now automatically retries with 50% layers, then 25%, and finally falls back to CPU mode to ensure a successful load.

🔌 Developer & API Enhancements

  • Enhanced Local API: The built-in REST API is now more robust for developer integrations (VS Code Continue, custom scripts).
  • Custom Model Icons: Local and custom-added models now feature a standard KaiROS logo and unified metadata display.
  • Improved Documentation: Updated README.md and GitHub Pages with detailed developer guides and high-resolution screenshots of the new RAG and Catalog features.

🛠️ Internal Improvements

  • Package Updates: All LLamaSharp backends (CUDA 12, CPU, Vulkan) updated to 0.25.0.
  • UI Polish: Added high-quality organization logos and model badges for better scannability.
  • Robustness: Improved error handling for LoadWeightsFailedException with detailed logging.

🔜 Coming Soon

  • Microsoft Store: The app is currently being prepared for its official Microsoft Store debut!

Run powerful LLMs locally with KaiROS AI — Private. Fast. Free.

What's Changed

  • feat: implement LLM model management, including a model catalog, cust… by @avikeid2007 in #11
  • New feature and design enhancement for the model catalog, hardware detection, and user experience by @avikeid2007 in #12
  • Add Chat Interface section to README with screenshot by @avikeid2007 in #13

Full Changelog: 1.0.4...1.0.5

Contributors

avikeid2007
Assets 3
Loading