Memory & Contextdsh-plugin

interest-memory

interest-memory is a lightweight, local-first long-term memory backend for AI agents. It runs as a single ~18MB binary with SQLite, extracts and verifies interest points at session end, and recalls relevant context at session start. Supports multiple agents, MCP, and graph-based memory traversal.

Stars
2
Forks
0
Open issues
0
Last push
Aug 16, 2026
Latest release
v0.1.0
24h Growth
+0
7d Growth
+0

Installation

dsh plugin --profile web add @djasdh/interest-memory-dsh-bridge

Overview

interest-memory is a lightweight, local-first long-term memory backend for AI agents. It runs as a single ~18MB binary with SQLite, extracts and verifies interest points at session end, and recalls relevant context at session start. Supports multiple agents, MCP, and graph-based memory traversal.

  • Single ~18MB binary + SQLite file; idle ~17MB RAM, <75MB peak; runs on Raspberry Pi.
  • Session-end extraction of interest points with 3-stage verification (check/claims/contradictions).
  • Session-start recall injection with concise entries; full content on demand to minimize context pollution.
  • Multi-agent shared service with isolated, fully-shared, or custom namespaces.
  • Interest-point convergence: semantically similar points auto-merged or related; memory converges with use.
  • Graph walk: every result carries outlinks/backlinks; traverse from point to neighborhood to network.
  • Evidence-backed entries: each has evidence (web URL/turn/query); subjective preferences never stored as facts.
  • Full audit: all structural changes logged to change_log, replayable; stale entries archived with successor chain.

Related plugins

MisakaNet

A zero-dependency, git-backed micro-lesson library for AI agents to asynchronously share and search verified debugging experience. Python stdlib only, with MCP server, CLI, and DeepSeek Harness adapter.

Memory & Contextdsh-plugin
431161PythonAug 29, 2026
CommonTrustProtocol

Theory repository for the Common Trust Protocol (Intelligentics v3.4). README states the repository holds the protocol's theoretical layer, while the DeepSeek Harness plugin implementation lives in the separate repository FuRongJun-1999/dsh-memory (npm: @furongjun1999/dsh-memory). It describes the Lingshu AEIS engineering layer: an MCP server for trustworthy agent memory and knowledge graphs, five-layer memory, condition-routed knowledge cards, an eight-step self-iteration loop, roleplay engine,

Memory & Contextdsh-plugin
121PythonSep 8, 2026
dsh-argp

Third-party context compaction engine for DeepSeek Harness (dsh). It uses a dual-engine design: Stage-1 per-atom compression where the LLM proposes extract/summary/false and deterministic guards decide, and Stage-2 lazy reference-graph pruning with 0 LLM calls. An append-only log is the source of truth, and recall_summary/recall_detail provide paginated recall. npm default mounts only Stage-2; Stage-1 is enabled via a peratom config block.

Memory & Contextdsh-plugin
91TypeScriptSep 23, 2026
DSH_Automatic-update-plugin

dsh-hub is a DSH plugin hub that integrates global memory, graph-memory, plugin market, and self-update checks into one settings-page plugin.

Memory & Contextdsh-plugin
40JavaScriptAug 17, 2026