mental-models
Staleness-Aware Memory: When Your Agent Should Verify Before It Trusts
Hindsight's consolidated memory (observations and mental models) is derived from raw facts, so it can lag behind new writes. Instead of silently serving a stale snapshot, Hindsight flags how far behind each layer is, and the reflect agent verifies against raw facts before trusting it.
How oh-my-pi Built Persistent Codebase Memory on Hindsight
oh-my-pi — a 10k-star terminal coding agent — uses Hindsight as its long-term memory backend. Here's a walkthrough of the integration: bank scoping, mental-model seeding, and the auto-retain pipeline, with code straight from the repo.
How Hindsight Learns: A Deep Dive Into Mental Models
Mental models are the top tier of Hindsight's learning hierarchy — persistent, self-refreshing understandings of the topics your agent asks about every session. A walkthrough from the schema up.