10Nates/Humanlike-AI-Chat

Employs a two-stage generation process with character-driven system prompts and a "decoherence" step to increase output entropy and human-like variation, running locally via Ollama integration. Supports temperature tuning and background generation to produce responses with higher stylistic variance than standard LLM outputs. Built as a Python CLI that wraps open-source models like Dolphin-Mixtral, enabling researchers to experiment with detection evasion techniques without external API dependencies.

No commits in the last 6 months.