dsh-plugin-asmemory

Most agent memory plugins store conversations or documents and answer the question “what did you say.” asmemory takes a different approach: it records two typed events—State (a value of an entity or metric at a point in time, e.g. gpu.temperature = 78°C) and Action (something that happened, e.g. agent started training)—and then answers “what happened and why” with time-series math.

Four deterministic analyses come out of the box: Trend (direction and slope of a metric), Anomaly (z-score outlier detection), Causal (mean change of a metric before vs. after an action, cleanly separating real causality from coincidence), and Summary (entity and event counts). Every result is pure arithmetic—no LLM guessing—and a single call exports a CSV ready for DataLens visualization.

It shines in three use cases: agent self-tracking (log its own actions and resource states, then ask why it got slow or hot), industrial monitoring (air-separation purity vs. valve opening, emission control), and personal data (sleep, weight, spending trends). If your agent needs to remember what happened in the physical or operational world rather than what was said in a chat, asmemory is purpose-built for that.