Apache-2.0 Rust MCP server + DeepSeek Harness plugin that gives agents a causal memory store surviving context compaction — facts, temporal state, and decision→outcome edges recalled by a hippocampus-style spreading-activation engine.
DSH integration
Ecosystem-related
Author-claimed
Safety audit
Unaudited
Last verified
2026-09-01
License
Apache-2.0
01What can it help you accomplish?
Give AI agents a causal memory that survives context compaction
A SQLite-backed store of decision → outcome causal edges with hippocampus-style recall, so agents stop re-making the same mistakes after compaction
Agent builders and engineers running DeepSeek Harness (dsh) who need long-term memory that compaction cannot erase
Stop agents from repeating past failures via causal intervention
An intervention_query that fires a DANGER chain citing a previously recorded lesson before a risky action
Teams whose agents take risky shell actions (e.g. `git push --no-verify`) and want guardrails drawn from prior lessons
02How to install into DeepSeek Harness
Not specified by the author
03DSH integration and capability boundaries
Ships as a DeepSeek Harness plugin (dsh-plugin/) with an MCP server backend
Causal memory store (decision → outcome)
agent decisions and their observed outcomes→decision → outcome causal edges persisted on one SQLite store, surviving compaction
Hippocampus-style spreading-activation engine
stored facts, temporal state, and causal edges→recall of what happened, when, why, and the counterfactual 'what would happen if'
Inhibitory causal edges (models inhibition)
causal relationships→inhibitory as well as excitatory causal edges — the README claims it is the only memory system that models inhibition
04Who is it for? When not to use it?
Good for
- Agent builders and engineers running DeepSeek Harness (dsh) who need long-term memory that compaction cannot erase
- Teams whose agents take risky shell actions (e.g. `git push --no-verify`) and want guardrails drawn from prior lessons
05Compatibility, maintenance and safety notes
Apache-2.0 · README advertises v0.9.2-alpha (release v0.9.2, 368 tests); no safety audit claimed
06Frequently asked questions
How does causal-memory connect to DeepSeek Harness?
The repository ships a dedicated DeepSeek Harness plugin — the README carries a 'DSH plugin' badge linking to the dsh-plugin/ directory — and causal-memory also runs as an MCP server, so DeepSeek Harness can use it through MCP or the bundled DSH plugin.
Does causal memory survive context compaction in DeepSeek Harness?
Yes. The causal table lives outside the agent's context window, so compaction cannot touch it. The README's benchmark shows 100% causal-table recall after 5 compactions, versus 45% for plain text.
What does causal-memory actually store and recall?
It stores facts, temporal state, and decision → outcome causal edges on one SQLite store, recalled by a hippocampus-style engine using typed spreading activation (excitatory and inhibitory), Hebbian reinforcement, Q-value dynamics, and immutable SWR consolidation.
Can causal-memory stop my DeepSeek Harness agent from repeating past mistakes?
Yes. An intervention_query can fire a DANGER chain that cites a lesson the agent recorded earlier — for example, warning before a risky `git push --no-verify`.
07Related DSH workflows
weknora
by tencent
Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.
mirage
by strukto-ai
The World's First Unified Virtual Filesystem For AI Agents
reme
by agentscope-ai
ReMe: Memory Management Kit for Agents - Remember Me, Refine Me.
memsearch
by zilliztech
A persistent, unified memory layer for all your AI agents (e.g. Claude Code, Codex, DSH), backed by Markdown and Milvus.
08Data and sources
[](dsh-plugin/)
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-09-01. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.