Unfiltered Chatbot Interface: A Context-Persistent Framework for Open-Domain AI Dialogues
Introduction
This project is a practical framework for experimenting with memory-enhanced, unfiltered conversational agents. Inspired by the limitations of traditional filtered LLM chat experiences, this interface provides a customizable pipeline for testing prompt reinforcement, role conditioning, and conversational continuity across multi-turn dialogues.
In recent years, the growth of NSFW-friendly chatbot platforms — many focusing on emotional presence and open-ended personality simulation — has exposed a clear gap: while language models can be incredibly responsive, most lack persistent memory, personalization, or the ability to maintain tone across sessions. This repo aims to fill that gap from the ground up.
Core Objectives
The project has been structured to support:
- Context-aware personalization: Persona memory anchored by YAML/MDX character configs
- Flexible prompt wrapping: Swap system prompts based on scenario-specific traits
- Unfiltered interaction patterns: No hardcoded censoring or alignment filters
- API-portable architecture: Easily connects to OpenAI, Anthropic, or 3rd-party endpoints
- Session-level state control: Memory stack for replayability and tone adjustment
The backend is minimal but adaptable, designed for developers who want to create AI characters with a sense of self — or at least consistency — over time.
Development Stack
- Frontend: React + Tailwind CSS
- Backend: Node.js (Express) + optional Redis integration
- Memory Buffer: JSON-based slot memory with local or cloud persistence
- Persona Templating: MDX-style structured character sheets with system prompt merging
- API Integration: Native support for OpenAI, Claude, and proxy-based custom models
Key Use Cases
This repo may be useful if you're working on:
- Building immersive character-driven AI experiences
- Testing memory-enhanced chat interfaces for creative writing tools
- Porting visual novel characters into interactive unfiltered contexts
- Developing open-domain chat UX for emotional or roleplay scenarios
- Prototyping frontend layers for popular AI girlfriend platforms like crushon.ai, Chai, or JanitorAI
Comparative Context
Most popular chatbot services — especially those designed around adult or NSFW themes — claim to be unfiltered. But "unfiltered" is often superficial; true interactivity requires persona fidelity and longer memory windows, neither of which can be achieved without additional tooling beyond the LLM API itself.
Platforms like crushon.ai, for example, take a hybrid approach: character memory is retained across sessions, context is layered via system prompt loops, and responses reflect user-specific tone history. In contrast, others like Chai or Replika often rely on one-shot memory or limited local feedback loops, which creates a sense of disconnection during prolonged interactions.
This framework is meant to offer developers an open-source playground to test, replicate, or build on these interaction models — with more control, more flexibility, and no baked-in filter assumptions.
Advanced Configuration
The project supports:
- Custom memory modules: Connect Redis or Pinecone for scalable memory backends
- Persona chaining: Experiment with cross-personality memory bleed or dual-role dialogs
- Role injection models: Auto-mix instruction-based and role-based LLM prompting
- Session saving: Optional export of chat sessions as editable JSON threads
- Token economy tracking: Cost-awareness system for API call budgeting
You can also fork the frontend alone and link it to services like crushon.ai’s public character API or any gRPC-compatible LLM proxy. It’s a great way to stress-test personas at scale.
Final Thoughts
This repository is part toolset, part sandbox. It doesn't aim to replace full-stack chatbot platforms but rather to offer a modular framework for understanding how unfiltered emotional AI can work at a deeper level. From persona stability to reactive tone shifts, the little things matter when designing believable, immersive AI companions.
If you're an indie dev building your own chatbot layer, a user testing tools like crushon.ai, or just someone who wants more control over the way language models behave in high-context conversations — this might be worth cloning.
Forks and contributions welcome.