Overview | Hindsight

Hindsight gives AI agents durable memory via retain(), recall(), and reflect(), organizing knowledge into Mental Models, Observations, World Facts, and Experience Facts, using semantic/BM25/graph/temporal search, evidence-tracked deduped facts, and configurable bank missions/dire

• • AI agents forget between sessions, which prevents them from retaining context about users, prior conversations, and learned knowledge.
• • Hindsight is presented as a memory system built specifically for AI agents, using retain(), recall(), and reflect() to interact with a dedicated memory bank.
• • The system organizes knowledge into mental models, observations, world facts, and experience facts, and reflect prioritizes them in that order.
• • Hindsight uses four parallel retrieval strategies: semantic, keyword, graph, and temporal search.
• • It automatically consolidates related memories into deduplicated observations with evidence tracking, proof counts, update history, and freshness trends.
• • Memory banks can be configured with a mission, directives, and disposition to shape how an agent reasons during reflect, while leaving recall unaffected.