v0.2.11

v0.2.11 - Fix GPU acceleration in installed builds

v0.2.10

Release v0.2.10 - Fix Release Build for GPU/CUDA

Fixes for GPU/CUDA support in installed builds:

1. cuBLAS Download Fix:
   - Changed URL to correct cudart-llama package (391MB)
   - cuBLAS DLLs are in separate package from main binary

2. CPU Backend Fix:
   - Added LLamaSharp.Backend.Cpu package for full ggml-cpu.dll
   - CUDA12 package has a stub that doesn't work
   - Workflow checks NuGet cache as fallback source

This should produce a working release with GPU acceleration.

v0.2.9

Release v0.2.9 - Fix GPU/CUDA Backend Loading

Fixes:
- GPU/CUDA now works correctly in installed builds
- Explicit backend loading via ggml_backend_load_all_from_path()
- NuGet CUDA12's stub ggml-cpu.dll replaced with full avx2 version
- Improved diagnostics for LLM initialization failures

This release fixes the issue where installed builds would fall back
to pre-canned emails instead of using GPU-accelerated LLM generation.

v0.2.7

v0.2.7 - Fix GPU: use CUDA-enabled DLLs from llama.cpp

v0.2.6

v0.2.6 - Add diagnostic logging for GPU detection

v0.2.5

v0.2.5 - Fix GPU detection in installed builds

v0.2.4

Release v0.2.4 - GPU Support & Model Improvements

Fixes:
- Fixed cuBLAS download URL in release workflow (was 404ing)
- GPU inference now works in installed builds (cuBLAS DLLs included)
- Removed TinyLlama model (low quality output)
- Changed CPU default to Qwen 2.5 3B

Improvements:
- Model download now prompts to restart game for GPU detection
- Switching AI models in settings shows restart notice
- Installer now cleans up config files on uninstall
- Added build-from-source PowerShell script for the paranoid

🤖 Generated with Claude Code

v0.2.2 - UI Fixes & Version Management

What's New

UI Fixes

• Fixed logo size in dashboard header to prevent overflow
• Window is now non-resizable in windowed mode (as intended)

Version Management

• Centralized version constant - game menu now shows correct version
• Version synced across game, export presets, and installer

Installer Improvements

• Now cleans up user config files on uninstall (settings, AI preferences)
• Prompts to delete downloaded AI models on uninstall (saves disk space)

---

Good work... for now.

v0.2.1

Release v0.2.1 - Branding Update

- InertiCorp logo replaces text titles throughout UI
- Custom app icon for Windows executable
- Consistent branding across dashboard, menus, and dialogs

- Dashboard header (left side)
- Game menu (pause menu)
- Settings menu
- Reports popup
- First-run setup dialog

Going Nowhere... Slowly. Now with style!

v0.2.0

Release v0.2.0 - Hardware-Adaptive AI & GPU Support

Major improvements to AI model selection and GPU support:

- Hardware detection on first launch (GPU vs CPU-only)
- Model choice dialog with Lite (TinyLlama) and Standard (Phi-3) modes
- Adaptive generation timeouts based on hardware and model tier
- InertiCorp logo added to dashboard and setup screen
- Custom app icon for Windows executable

- Fixed CUDA backend not loading properly in released builds
- Fixed native DLLs not being found in correct directories
- Added CPU fallback for mixed operations (.WithAutoFallback())

- Release workflow now copies native DLLs to cuda12 directory
- Multi-path search for llama.dll and ggml-*.dll
- GGML_BACKEND_PATH environment variable set for backend discovery

Going Nowhere... Slowly. But now with better AI support!