Overview | Hindsight

AI agents start each conversation from zero, so they forget prior context, user details, and learned knowledge between sessions.
Hindsight is a memory system built specifically for AI agents and exposes three main actions: retain(), recall(), and reflect().
Knowledge in Hindsight is organized into Mental Models, Observations, World Facts, and Experience Facts.
During reflect, Hindsight prioritizes sources in this order: Mental Models, then Observations, then Raw Facts.
Hindsight uses four parallel retrieval strategies: semantic, keyword (BM25), graph, and temporal search.
Temporal search is meant for questions like “What did Alice do last spring?” where time reasoning matters more than similarity.
The system can connect related facts, such as using “Alice works at Google” and “Google is in Mountain View” to answer related location questions.
Hindsight consolidates facts into deduplicated observations and tracks evidence, proof counts, update history, and freshness trends.
Memory banks can be configured with a mission, directives, and disposition to influence how reflect reasons.
These bank settings affect reflect but do not change recall.

Your notes

Save this item to your library to add private notes.