memory-kernel

You persist AI agent memory across sessions by storing structured, typed atoms as markdown files with metadata like type, status, and confidence. This prevents forgetting and drift by using files as the source of truth for reliable recall.

Typed memory for AI agents is a structured store that preserves knowledge like decisions, constraints, and beliefs using a defined schema with atom types and confidence levels. You need it for long-running tasks requiring reflection and rule enforcement.

You detect conflicting memories by using a memory kernel that auto-surfaces conflicts during recall. It maintains consistency by evaluating typed atoms and supporting end-of-session cleanup to resolve contradictions in stored knowledge.

Yes, you can use an optional SQLite index for fast recall queries in long-running AI agents. The typed markdown files remain the source of truth, while SQLite provides quick retrieval and filtering over the stored atoms.

Yes, AI agent memory persistence works without a database by using git-friendly markdown files as the source of truth. The optional SQLite index enhances query speed but is not required for basic typed memory storage and recall.

The best way to store AI agent beliefs and constraints is using typed memory atoms with a defined schema. Each atom includes a type, status, confidence value, and optional TTL, allowing precise evaluation of rules and reflections across tasks.