v0.2.4

What's new

• Added a Voice Input tab. It installs, updates or removes local speech to text (the whisper.cpp engine and an English model) so you can talk to survivors in the game. Every download is checked against a pinned SHA-256 hash. Choose Fast (141 MB, default) or Accurate (181 MB, about 3x slower).
• Remnants LLM is now recommended. The LLM Companion tab has a "Use Remnants LLM" button that selects Remnants LLM v5 (hosted on Hugging Face as PatrickPat/remnants-llm-v5-GGUF, about 2.5 GB). It is listed first in the model catalogue, is the default choice for new users, and the tip line says whether it fits your GPU.

Download

• The attached Windows x64 executable is self contained, so no separate .NET installation is needed.
• It is not code signed, so Windows Smart App Control or SmartScreen may warn about it.
• SHA-256: 77d8bd3ce30b80070152e73ec2451b4fbc83d599e16f4f735ade975eb20038fc

v0.2.3

What's new

• Added Benchmark selected model to the local Ollama tab, with an embedded versioned pack of 223 forced-model cases.
• Every case sends a real Ollama request
• Includes progress, cancellation, category scores, per-case expected and actual actions, failure reasons, and an exportable detailed report.
• Reports cold-load and response timing, token throughput, context, quantization, and GPU/CPU placement where Ollama provides them.

v0.2.2

What's new

Yeah... so basically it warms up the model first

Download

• The attached Windows x64 executable is self contained,so no separrate .NET installation is needed.

v0.2.1

fix for some rx cards

v0.2

Model catalogue, GPU detection and remote model lists.

• Browse and download Ollama models in-app, labelled by what your GPU can actually run
• Correct model recommendations based on real VRAM
• OpenAI-compatible endpoints can list their available models
• Reworked LLM panel: GPU tip, scrollable log, per-provider model lists
• MIT license

Self-contained win-x64 build, no .NET runtime required.

v0.1

Initial Project Remnants Setup release.