Release PrivateLM App v1.0.4 · orailnoor/cross-platform-llm-client

Document AI — Upload and analyze PDF and DOCX files. Text is automatically extracted and added to the conversation context.
Image Generation Progress — Added a real-time progress bar with percentage, step counter, and estimated remaining time during image generation.
Device Compatibility Advisor — Settings now detects your device chipset (Snapdragon, Tensor, Apple Silicon, etc.) and recommends the most suitable model quantization format.
Resume Session Dialog — On startup, the app now asks whether you want to reload previously used models instead of restoring them automatically.
Local Mode by Default — Fresh installs now default to on-device inference instead of cloud mode.

Fixed chat context leaking between conversations when switching chats.
Fixed incorrect “restart required” warnings when switching from LiteRT models to GGUF models.
Fixed empty responses on Pixel 6, 7, and 8 devices by running Gemma models in single-threaded mode on Google Tensor chips, along with automatic garbage-token filtering and clearer error handling.
Fixed temporary vision attachment files not being cleaned up after removal.
Fixed cross-chat message contamination when deleting or switching chats during active AI streaming.
Fixed multiple memory leaks across Server, Task, and Model views involving TextEditingController and StreamSubscription instances.

Improved the New Chat button with a more intuitive edit_note icon and tooltip.
Replaced raw native image model crash dumps with actionable recommendations such as trying CyberRealistic, Realistic Vision, or AbsoluteReality.
Unsupported file formats are now blocked with a clear toast message instead of failing silently.
Added active image model indicators in the app bar.
The last used image model is now remembered across app restarts.

SDXS 512 — Removed from the image model catalog due to widespread compatibility issues across devices.

Provide feedback