graph-memory: A Knowledge Graph Memory Plugin for DeepSeek Harness
adoresever/graph-memory
Graph Memory is a native knowledge-graph memory plugin for DeepSeek Harness, extracting structured triples from conversations, compressing context by ~75%, and enabling cross-session experience reuse with traceable recall.
graph-memory is a DSH plugin for DeepSeek Harness that turns conversation history into traceable, searchable, cross-session memory as a knowledge graph. By extracting structured triples like TASK, SKILL, and EVENT with typed edges, it injects only relevant local subgraphs during recall, compressing context and reducing token usage. It offers native host integration, durable memory, smart recall, and observability, all local-first without requiring a graph database.
How to Install
dsh plugin --profile web add github:adoresever/graph-memory- Category
- Memory & Context
- Platform
- DSH Plugin
- Author
- adoresever
- Distribution
- Plugin
graph-memory Key Features
graph-memory Repo Summary
What Does It Do?
Graph Memory is a DSH plugin for DeepSeek Harness that turns conversation history into traceable, searchable, cross-session memory as a knowledge graph. It solves the problems of context bloat and lost experience across sessions by extracting structured triples (like TASK, SKILL, EVENT) and typed edges (like USED_SKILL, SOLVED_BY), so recall only injects a relevant local subgraph instead of replaying the full history. Core capabilities include native host integration, durable cross-session memory, context compression (measured around 75% token reduction in a specific workflow), and a local-first lightweight deployment.
Core Features
- Native integration: loaded via the DSH/Cordis plugin lifecycle, hooking into Session, Tool, Agent Loop, Prompt Assembly, LLM, and Credentials seams without forking or modifying the DSH core.
- Durable cross-session memory: knowledge from Session A can be recalled in Session B, memory survives DSH restarts, and stable event IDs make ingestion idempotent.
- Smaller, cleaner context: keeps the newest real user turns verbatim (default 5), uses the DSH compaction service to replace older prefixes with a rolling checkpoint, and indexes each checkpoint with provenance.
- Smart recall: combines semantic vector retrieval with FTS5 lexical fallback, uses a high-precision semantic gate (default 0.6) for automatic injection, and marks recalled history as untrusted reference material.
- Observability: gm_status reports store path, graph counts, vector coverage, mode, and dimensions; re-embedding triggers on model or dimension changes.
How to Use This Plugin?
After enabling it in DSH, the plugin automatically processes conversations and builds the knowledge graph. You can explicitly record critical knowledge with gm_record or perform broad searches with gm_search. In daily use, no manual intervention is needed—the plugin recalls relevant history and injects it into the current prompt automatically. To adjust behavior, modify configuration options such as freshTurnCount (number of raw turns to keep), recallTokenBudget (default 4096), and autoRecallMinScore (default 0.6).
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: adoresever/graph-memory. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.