Back to Documentation
Memory Types
5 structured categories of AI memory: conversation, knowledge, learning, decision, architecture.
Memories stored by LOB Brain are partitioned into structured categories to ensure precise, predictable retrieval by LLM Agents.
- knowledge: Facts, standards, and static logic. Highly reusable codebase reference materials.
- architecture: High-level design schemas, system infrastructure, Architecture Decision Records (ADRs), and module boundaries.
- decision: Critical overrides and architectural choices agreed upon by humans rather than algorithms (e.g. "Use App Router over Pages Router to eliminate transition delay").
- learning: Post-mortem takeaways captured after resolving complex bugs (e.g. "Fixed connection timeout. Avoid binding to wildcard address in production").
- conversation: Summarized conversation logs between users and agents. Subject to rapid decay to keep the database lightweight.