hindsight vs metronix-memory
Agent memory that learns, not just recalls: retain/recall/reflect API over Postgres, SOTA on LongMemEval. Self-host via Docker with UI; Python/TS clients, any LLM provider. — versus — Self-hosted agent memory stack in Docker: Postgres + Qdrant + Neo4j hybrid retrieval, a temporal knowledge graph, ontology layer and freshness checks behind one MCP-native API.
Same job — self-hostable agent memory benchmarked on LongMemEval. Hindsight bets on reflect-and-learn over Postgres; Metronix bets on retrieval breadth (dense + sparse + graph). Pick by whether your failure mode is recall or reasoning over what was recalled.
| hindsight | metronix-memory | |
|---|---|---|
| Stars | 22k | 95 |
| Forks | 1.7k | 9 |
| Language | Python | Python |
| License | MIT | Apache-2.0 |
| Last activity | today | 3 days ago |
| Topics | memory, agents | memory, rag, knowledge-graphs |
| Curated connections | 8 | 6 |
hindsight — the curator's take
Pick it when you want a deployable memory *service* whose pitch is learning — agents that get better over time, not a transcript search. The LongMemEval lead was independently reproduced (Virginia Tech, Washington Post), which is more than most memory vendors offer, and the LLM side is pluggable down to Ollama/LM Studio for fully-local stacks. NOT an embedded library: you run a Docker service with Postgres and talk to it over HTTP — overkill for a single coding agent wanting session notes. The ™ and Hindsight Cloud signal a commercial trajectory; watch where the open/paid line lands.
metronix-memory — the curator's take
Reach for it when you want one self-hosted box doing dense + sparse + graph retrieval and keeping long-lived facts fresh, and you're willing to run four datastores to get it. Skip it if you only need per-project session recall — a SQLite memory plugin is a fraction of the operational surface. Read the benchmark table with care: they are the author's own N=1 runs under a self-defined protocol, and the honest signal in them is that retrieval scores (Recall@10 85-95%) far outrun end-to-end answers (53-63%) — finding the evidence was never the hard part.