soul-of-waifu

Soul of Waifu (SoW) is a free, open-source Windows desktop application where AI characters exist across multiple interconnected interaction modes simultaneously. Created characters maintain a single continuous identity across text and voice chat, a tabletop RPG engine, and an autonomous desktop agent. Core modules include: 1. Soul Memory: A long-term cognitive architecture that tracks mood, relationships, and lived history through four isolated cognitive files (Psychology Layer, Relationship Profile, Episodic Topic Archive, Personal Diary). Features emotional decay over time, memory self-healing for contradiction detection, duplicate detection, and crash-safe atomic writes. Offers four depth modes (Full Sync, Soul Link, Mind Spark, Reflection Flow) to fit different VRAM/RAM constraints. 2. Text & Voice Chat: Full-duplex conversation with Live2D and VRM avatars, customizable state variable HUD, lorebooks, and real-time voice calls. Supports 6 TTS engines (Qwen3 TTS, XTTSv2, Kokoro 82M, Silero, EdgeTTS, ElevenLabs), RVC voice conversion, sentence-by-sentence streaming, and a 28-emotion classifier with LipSync. 3. Soul Stage: A tabletop RPG simulator with a dedicated AI Game Master. Includes deterministic dice rolls and skill checks, hidden Story Arcs, a Campaign Board with pressure clocks, structural relationship tracking, turn control modes (Say/Do/Think/Director), private whispers, inventory and status HUD, scene folders, atmospheric automation, and event cards. 4. Soul Companion: An autonomous desktop agent with neurohormonal simulation (dopamine, cortisol, oxytocin, fatigue). Can read the screen, control mouse/keyboard, browse the web via headless Playwright browser, run sandboxed Python/PowerShell scripts, organize files, and chain multi-step tasks. Extensible via custom Python plugins and MCP servers. All risky actions are gated behind a Human-in-the-Loop approval banner with a 25-second fail-safe countdown. Additional features include custom state variables with 11 built-in presets (Romance, RPG, Survival, Horror, Visual Novels, etc.), a local web client for mobile/tablet access over Wi-Fi, a Discord gateway with multi-user awareness and computer vision, and AI image generation integration (Automatic1111, ComfyUI, NovelAI, DALL-E 3, FLUX). LLM backend support includes local Llama.cpp with CUDA, HIP, SYCL, and Vulkan acceleration, plus 10 cloud providers (OpenRouter, OpenAI, Claude, DeepSeek, Grok, Gemini, Qwen, Mistral, Z.AI, Player2). Advanced text sampling options include dynamic temperature, DRY sampler, XTC, and Min-P. The integrated Models Hub provides Hugging Face search, GGUF downloads, and automated Llama.cpp updates. The application runs entirely locally by default with no account requirements. Cloud AI providers are optional. Installation requires no Python or command-line knowledge. Licensed under GPLv3.