🎭 DEF(kari) — Persistent Character Platform

Dialogue × Emotion × Fable
With your characters, for years to come, wherever you go.

DEF(kari) is a local-first AI creative platform that integrates text, voice, and image generation.

Rather than entrusting your creative content to cloud service terms and content policies, DEF(kari) provides the foundation for generating and sustaining the characters and stories you desire, in your own environment, with your own hands.

No GPU? You can still get started.

DEF is local-first, but you don't need a local environment to begin. Using external APIs (Gemini / OpenAI / Anthropic), text generation and voice synthesis work without a GPU. Image generation requires a T2I API (OpenAI DALL-E / Civitai / Hugging Face) or a local GPU environment. Once your local setup is ready, switch to fully offline operation at any time.

GitHub Repository

Screenshots

Chat Mode

Session Mode

Novel Mode

TRPG Mode — Session Setup

TRPG Mode — Session in Progress

Character

Key Features

Local-First — LLM, TTS, T2I all run locally. External API fallback supported
No GPU required to start — Text + voice via external APIs. Switch to local GPU anytime
3-Modality Integration — Text, voice, and image as a continuous creative experience
4 Modes — Chat / Session / TRPG / Novel
Character Persistence — Resume from where you left off
Adapter Pattern — 5 LLM, 5 TTS, 5 T2I backends
OpenAI-Compatible API — Connect any compatible service: LM Studio, vLLM, llama.cpp, Groq, OpenRouter, and more
Zoning — Public / Private data separation

Backend Support

Layer	Local (GPU)	External API (no GPU)
LLM (text)	Text Generation WebUI / Ollama	Gemini API / OpenAI API / Anthropic Claude API / Groq / Grok / OpenRouter
TTS (voice)	VOICEVOX / Kokoro TTS / Irodori-TTS	Gemini TTS API / OpenAI TTS API / Grok TTS
T2I (image)	Automatic1111 / ComfyUI	OpenAI DALL-E API / Civitai API / Hugging Face API

Note: DEF(kari) is designed to run on your local machine, not as a Hugging Face Space application. Visit the GitHub repository for installation, documentation, and user guide (JP/EN).

Tech Stack

Python / FastAPI / React (TypeScript) / AGPL v3