Xplore-LAB/dsh-plugin-asmemory
Action-State Memory Engine: typed time-series memory (states + actions) with trend/anomaly/causal analysis for
Gives an agent a time memory instead of a text memory. asmemory stores two kinds of typed events — State (a value of some entity or metric at a point in time, such as gpu.temperature = 78C) and Action (something that happened, such as an agent run or an operator adjustment) — and then answers four questions on top of them: trend (slope and direction of a metric), anomaly (z-score outliers), causal (the before/after delta that attributes a metric move to an action) and summary (entity and count inventory). The README frames it as the memory layer for the physical and operational world: agents observing themselves, infrastructure and lab work, rather than the record of what was said.
Install
pip install .
dsh web --patch "$PWD/cordis.yml"README's English install blocks quoted verbatim: the Python package is installed from the checkout and the plugin is then booted by pointing dsh web at the repo's cordis.yml. The README adds that once the package is published you will also be able to run dsh plugin add dsh-plugin-asmemory; registry check 2026-09-17 returned 404 for dsh-plugin-asmemory, so that form is not yet available and this page uses the source route.
Compatibility
README: a DeepSeek Harness plugin whose engine is a Python package installed from the checkout, booted through the repo's cordis.yml patch overlay. The README publishes a 'Verified' section for its tools and a quick-start example.
Details
- Repo: Xplore-LAB/dsh-plugin-asmemory
- Category: Agent Capabilities
- Stars: 0
- Version: Source install (no registry version asserted; repo 0★, MIT, last push 2026-08-14)
- Last push: 2026-08-14
- First seen: 2026-08-13
Recent updates
The README documents the event model, the four analyses, the tool list and the install route rather than a release-by-release table.
FAQ
- How do I install it?
- The README's route is pip install . from the checkout and then dsh web --patch "$PWD/cordis.yml"; the README notes a published dsh plugin add dsh-plugin-asmemory form will exist later, and that npm name did not resolve on 2026-09-17.
- How is this different from other memory plugins?
- Per the README, most memory plugins store conversations or documents and answer 'what did you say'; asmemory stores actions and states and answers 'what happened, and why' — trend, anomaly and causal questions over time series.
- What does it give the model?
- The README lists typed State/Action recording plus four analyses — trend, anomaly, causal and summary — exposed as plugin tools with a documented example workflow.