rlat-learn

What problem does it solve?

It guides an rlat knowledge model to extract and store durable world facts, standing constraints, and learned findings from in-session user statements, with strict privacy safeguards.

Core Features & Use Cases

• Four gates framework: enforces durability, user-statement provenance, premise relevance, and world-only scope to filter candidates.
• Land-and-confirm loop: prompts user confirmation before landing each fact into the knowledge model via rlat capture-attribute.
• Scoped in-session learning: captures only world-relevant information and prevents personal data leakage, supporting iterative updates during a session.

Quick Start

When the user says remember this or rlat should know this, mine and land approved facts into the knowledge model after user confirmation.