ModikDialog

ModikDialog

LLM trader dialogues for Casualties Unknown

Description

A BepInEx plugin by Staili for Casualties Unknown single-player trader dialogue. Right-click any trader — get a custom menu instead of instant barter.

Talk to traders with a local AI model: replies use your reputation, their stock, what you're holding, and your current state. No cloud API, no API key — everything runs on your PC via llama-server (llama.cpp) + Qwen2. 5-3B-Instruct GGUF (~2 GB, manual download).

Main features

• []Trader menu — right-click a trader: Talk / Trade / Leave (RU: Общаться / Торговать / Уйти). Trade opens the vanilla barter window.[/]

  []LLM dialogue — Talk starts a conversation: greeting, speech bubbles, clickable question buttons. Each answer is generated locally.[/]

  []Game context — the model sees trader reputation, inventory, item in hand, and player state for relevant replies.[/]

  []Languages — RU / EN auto-detected from game locale (no manual language setting).[/]

  []Performance — profiles Fast / Balanced / Quality in settings.conf. CPU-only, no GPU required (AVX2, 4+ cores recommended).[/]

Installation

• []Install BepInEx 5. 4.x — winhttp.dll next to CasualtiesUnknown.exe[/]

  []Install CUCoreLib and QoL. Unknown → BepInEx\plugins\[/]

  []Create BepInEx\plugins\ModikDialog\ and copy ModikDialog.dll there[/]

  []Download llama.cpp (llama--bin-win-cpu-x64.zip) — extract into BepInEx\plugins\ModikDialog\models\llama\ (llama-server.exe + DLLs)[/]

  []Download Qwen2. 5-3B-Instruct Q4_K_M (~2 GB) → BepInEx\plugins\ModikDialog\models\qwen2. 5-3b-instruct-q4_k_m.gguf[/]

  []Right-click a trader in-game — the custom menu appears on the left[/]

  []Settings: BepInEx\plugins\ModikDialog\models\settings.conf (fallback: BepInEx\config\ModikDialogByStaili.cfg)[/]

• []Required mods (Nexus):

  []CUCoreLib — CUCoreLib.dll in BepInEx\plugins\[/]

  []QoL. Unknown — QoL Unknown.dll in BepInEx\plugins\[/]

[]Required (non-mod): Casualties Unknown Demo + BepInEx 5. 4.x (Windows x64)[/]

[]Required (non-mod): llama-server + Qwen2. 5-3B-Instruct GGUF (~2 GB) — manual download, mod does not fetch files[/]

[]Required (non-mod): CPU 4+ cores (AVX2), ~2. 5 GB free disk[/]

Notes

• []First reply may take a few seconds while llama-server loads the model[/]

  []If slow: set Profile = Fast in settings.conf, or use Qwen2. 5-1. 5B (lighter model)[/]

  []Check BepInEx\LogOutput.log if llama-server.exe or the GGUF file is not found[/]

  []Optional: QoL. Unknown[/]

  ---

  Support the author

  • []DonationAlerts (RUB, USD, EUR…): donationalerts.com/r/steyferstaili[/]

    []TON / USDT: UQCI4s0ImjWashSanglRuIJuCIsjlJfSPedVR1YSUa45q4yS[/]

    []Telegram: @Stailick

    ---

    Shout outs

    Thanks to the Qwen team, llama.cpp contributors, BepInEx maintainers, and the Casualties Unknown community.

    Bug reports — Posts tab or Teleg