Repolode
Submit
adoresever

adoresever/graph-memory

A knowledge-graph memory plugin for DeepSeek Harness and OpenClaw that bounds recent context and recalls exact past Q/A across sessions.

View on GitHub
Stars
661
Forks
98
Open issues
21
Last commit
22 days ago

What is it?

Graph Memory is a memory plugin for DeepSeek Harness (DSH) that is also compatible with OpenClaw. It keeps the most recent completed user turns in the model's context, archives older history into a single marker, and removes completed reasoning and tool traces from future requests. The full event log is preserved locally.

After each completed turn, the plugin writes a short summary and extracts subject-predicate-object triples from it. These form a graph used for navigation: when a question matches older or cross-session material, the plugin recovers the original user question and final answer as evidence. Retrieval uses vector Top-K search with an FTS5 full-text fallback, and storage is a local SQLite database.

The project reports a 20-turn benchmark run in its README, where first-request context on turn 20 dropped from about 57k to about 11k tokens. Extraction can use a dedicated OpenAI-compatible or configured model, and embeddings are optional.

Key features

  • 01Keeps the newest completed turns in context and archives older history into one marker.
  • 02Stores a per-turn summary and SPO triples in a local SQLite graph.
  • 03Recalls exact source question and answer when older or cross-session memory matches.
  • 04Falls back to FTS5 full-text search when no embedding endpoint is configured.
  • 05Supports OpenAI-compatible embedding endpoints and a dedicated extraction model.
  • 06Provides DSH tools for status, search, recording, and retry of failed extraction.

Use cases

  • Long-running coding or research sessions in DeepSeek Harness where the context window fills up.
  • Carrying decisions and facts from earlier sessions into new conversations.
  • Running an OpenClaw agent that needs compact, searchable conversation memory.

Quick start

  1. 1

    Install the pinned plugin into DeepSeek Harness

    npx @deepseek-ai/dsh plugin --profile web add github:adoresever/graph-memory#v1.6.0-beta.17
  2. 2

    Check the resolved configuration

    npx @deepseek-ai/dsh --profile web --dump-config
  3. 3

    Start the DSH web interface and confirm graph-memory/dsh is active under Settings → Plugins

    npx @deepseek-ai/dsh web

Things to know

  • Targets DeepSeek Harness, and the npm registry still serves an older 1.5.8 release, so install from the pinned GitHub release.
  • Requires Node.js 22.13 or newer.
  • Current version is a beta (1.6.0-beta.17).
  • Semantic recall without embeddings falls back to FTS5, and embeddings need a separate OpenAI-compatible endpoint and key.

In collections

Claude Code tools, skills & pluginsAI coding assistants & agentsDatabases & SQL tools

Topics

#knowledge-graph#memory#context-management#llm#sqlite#vector-search#typescript#claude-code#codex#deepseek

FAQ

What is graph-memory?
A knowledge-graph memory plugin for DeepSeek Harness and OpenClaw that bounds recent context and recalls exact past Q/A across sessions. Graph Memory is a memory plugin for DeepSeek Harness (DSH) that is also compatible with OpenClaw. It keeps the most recent completed user turns in the model's context, archives older history into a single marker, and removes completed reasoning and tool traces from future requests. The full event log is preserved locally.
Is graph-memory free to use? What license does it have?
graph-memory is open source under the MIT license. Check the license terms before using it in commercial products.
Is graph-memory still maintained?
Actively maintained. The last commit to graph-memory was on Oct 1, 2026.
What can I use graph-memory for?
Long-running coding or research sessions in DeepSeek Harness where the context window fills up. Carrying decisions and facts from earlier sessions into new conversations. Running an OpenClaw agent that needs compact, searchable conversation memory.
How do I install graph-memory?
Install the pinned plugin into DeepSeek Harness: npx @deepseek-ai/dsh plugin --profile web add github:adoresever/graph-memory#v1.6.0-beta.17 Check the resolved configuration: npx @deepseek-ai/dsh --profile web --dump-config Start the DSH web interface and confirm graph-memory/dsh is active under Settings → Plugins: npx @deepseek-ai/dsh web
Where can I get graph-memory?
The source code and installation instructions are on GitHub: https://github.com/adoresever/graph-memory

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects