I built this because generic LLMs are often too moralizing or biased when dealing with ethical non-monogamy (ENM) and polyamory contexts. I wanted a tool that acts as a "linguistic calibrator" to help partners communicate securely, specifically respecting attachment styles (Anxious/Avoidant).

The Tech Stack:

- Frontend: Next.js 14 (App Router), Tailwind

- Backend: Supabase (Postgres + pgvector for long-term memory of relationship graphs)

- AI: Model-agnostic architecture (currently routing via OpenRouter to GPT-4o and Sonnet 3.5). I use heavy system prompting to enforce NVC (Non-Violent Communication) principles.

The core challenge was preventing the context window from getting messy with complex polycule structures (who is dating whom), so I implemented a graph-based memory retrieval system.

Would love to hear your feedback on the prompts or the approach to privacy!