Causal Memory gives agents a memory graph with inhibition

Field notes

The project stores facts, temporal state, and causal edges together in a SQLite-backed memory system, then exposes intervention queries for hypothetical actions. Its published CausalEval snapshot covers 140 questions across 20 graphs, while the compaction table reports 100% causal-table recall at five compactions compared with 45% textual recall. Installation is described through pip or a Rust build, with an MCP configuration for agent clients.