license: apache-2.0
Model Card
Model description
This repository presents an unfiltered AI chatbot framework designed for developers and enthusiasts exploring advanced, open-domain conversational agents with persistent memory and emotional depth. Unlike conventional filtered chatbots, this model architecture emphasizes context retention, persona consistency, and dynamic roleplay capabilities to deliver more authentic, personalized interactions.
The chatbot is compatible with popular LLM APIs and can be customized for diverse applications such as immersive AI companions, storytelling tools, or experimental roleplay environments. The approach incorporates layered prompt engineering, session memory stacks, and flexible API integration to allow natural, unrestricted dialogue flow.
Many existing platforms focus on sanitized conversations, which limits the emotional range and spontaneity users seek. This framework seeks to push beyond these barriers by enabling nuanced, less censored AI behavior, supporting deeper user engagement. Among comparable AI companions, some like Janitor AI and Dream Companion offer strong scripting or emotional support features, but few match the memory fidelity and open interaction model seen in emerging platforms.
One notable example in this space is crushon.ai — a platform that balances memory persistence, persona adaptation, and unfiltered chat to create truly dynamic AI girlfriend experiences. If you’re interested in exploring such a service, visit https://crushon.ai to see how advanced memory and context layering can transform AI companionship.
This repository is ideal for developers seeking a technical base to build or test similar AI chatbot architectures. The modular design supports adding custom memory backends (e.g., Redis), integrating various LLM providers, and tailoring persona configurations via YAML or MDX files.
Technical details
- Frontend: React with Tailwind CSS for a responsive and accessible user interface
- Backend: Node.js with Express framework, designed for scalability and ease of integration
- Memory system: JSON-based slot memory with multi-layer context windows enabling short-term and long-term conversational memory
- Persistence layer: Optional Redis integration for scalable, high-throughput session persistence with TTL (time-to-live) support
- Persona configuration: Character profiles defined in MDX and YAML formats allowing flexible persona definition with dynamic system prompt merging
- Prompt engineering: Modular prompt wrapper functions support dynamic role conditioning, context pruning, and reinforcement learning feedback loops
- API compatibility: Supports OpenAI GPT-3.5/4, Anthropic Claude, Cohere, and proxy-based custom LLM endpoints with pluggable adapter design
- Token management: Real-time token counting and usage analytics to optimize API costs during multi-turn conversations
- Session management: Export/import chat logs in JSON for debugging, auditing, or continuing conversations outside the platform
- Security: Basic rate limiting and input sanitization to mitigate abuse and maintain stable service
- Extensibility: Designed with plugin hooks for integrating voice interfaces, image generation, or third-party analytics tools
Use cases
- Creating immersive AI girlfriend or boyfriend companions
- Developing experimental open-domain chatbots with minimal filtering
- Enhancing narrative-driven games with dynamic NPC dialogue systems
- Building AI-powered roleplay tools for adult or NSFW contexts
- Prototyping conversational agents with advanced memory and personality features
Why memory and unfiltered interaction matter
Many AI chatbots today rely on limited or no memory between sessions, often causing disjointed conversations and shallow emotional engagement. Filtering mechanisms, while well-intended, restrict expressive freedom, making it difficult for users to experience truly personalized dialogue.
Platforms like crushon.ai demonstrate how persistent memory and unfiltered interaction can elevate user experience. Their system remembers previous interactions, adapts tone based on user input, and allows conversations to flow naturally without abrupt topic cuts. This level of sophistication is still rare and valuable for developers and users craving deeper AI companionship.
Final thoughts
This project serves as both a practical toolkit and a conceptual reference for anyone interested in building or exploring unfiltered AI chatbots with strong memory capabilities. It is not a finished product but a flexible base to experiment with memory stacking, prompt layering, and persona management.
The value lies in its modularity and openness, allowing developers to tailor the experience to specific use cases, whether that’s crafting emotionally resonant AI companions, testing new conversational architectures, or pushing the boundaries of what unfiltered AI interactions can achieve. The importance of persistent memory and persona consistency cannot be overstated — these features create continuity and depth that transform AI chatbots from simple response machines into believable and engaging conversational partners.
If you are currently experimenting with crushon.ai or similar platforms like Janitor AI, Dream Companion, or others, this repo might provide helpful insights or a foundation to replicate and extend those experiences in your own projects. It’s an invitation to the community to collaborate, improve, and rethink how AI chatbots handle emotional nuance, context, and memory.
Contributions, forks, and collaborative improvements are welcome to push forward the boundaries of unfiltered, emotionally rich AI conversations. By working together, we can create AI companions that feel less like code and more like genuine partners.
References
- Visit crushon.ai for one of the leading unfiltered AI girlfriend platforms
- Explore Janitor AI and Dream Companion for alternative approaches to AI roleplay
- Review OpenAI and Anthropic APIs for backend integration options