DirectAI — Pre-A Release Devlog #1

This is the first public Pre-A release of DirectAI. It represents the first fully integrated build of the system, combining local inference, sandboxed execution, and bundled model runtime.

Model / Runtime Notes

The default model is an optimized 8B instruct model designed for efficient local inference on consumer hardware.

• Quantized for reduced memory usage
• Targeted for 16 GB RAM systems
• Runs on integrated GPUs via Vulkan (performance may vary)
• Dedicated GPUs significantly improve throughput

Known Issues

• Installer UI may become unresponsive if network interruptions occur during download
• Visual download progress can desync from actual backend state
• Despite UI issues, downloads continue in the background
• Application will automatically launch once installation completes
• Model size (~4–4.2 GB) can be used as a manual completion indicator

Notes

• UI instability is isolated and does not affect installation integrity
• No restart is required in case of UI failure
• Sandbox system is still in early development and will be expanded

This Pre-A release is considered experimental and subject to architectural changes.