We don't just chat.
We remember.
Standard chatbots reset when you close the tab. LibreMind is built on a persistent, dual-layer memory architecture designed for long-term care.
Recursive Memory Compression
Healing requires continuity. Our backend utilizes a sliding context window: recent messages are kept raw for immediacy, while older conversations are recursively summarized into vector embeddings. This allows the AI to recall themes from weeks ago without hitting token limits, mimicking human long-term memory.
Active Risk Guardrails
Passive listening isn't enough when safety is at risk. We implemented a real-time 'Safety Circuit' that scans every input before generation. If high-severity distress is detected, the system bypasses the standard LLM response to trigger specific safety protocols and flag the event for institutional review.
Row-Level Sovereignty
Privacy isn't just a promise; it's database law. We rely on Postgres Row Level Security (RLS) policies. This means the database engine itself rejects any query for your journal entries that doesn't originate from your authenticated session. Even our own admin dashboards cannot accidentally expose your raw thoughts.