Graph Memory

Compaction answers “how much of this conversation still fits?” Graph Memory answers “which past knowledge is worth recalling now?”

Reusable conversation knowledge becomes typed nodes:

• TASK: goals, execution, and outcomes;
• SKILL: validated reusable methods;
• EVENT: errors, fixes, decisions, changes, and facts.

Typed edges such as USED_SKILL, SOLVED_BY, REQUIRES, PATCHES, and CONFLICTS_WITH preserve relationships. A new question retrieves a relevant local subgraph instead of replaying the complete history.

Core advantages

Native host integration

• Loaded by the DSH/Cordis plugin lifecycle, not simulated through an MCP side channel.
• Integrates Session, Tool, Agent Loop, Prompt Assembly, LLM, and Credentials seams.
• Disposes database, cache, and event listeners with its plugin fiber.
• Does not fork or modify DeepSeek Harness core.

Durable cross-session memory

• Knowledge from Session A can be recalled automatically in Session B.
• Memory survives DSH restarts.
• Stable event IDs make resume and HMR ingestion idempotent.
• Source sessions and graph edges explain why a memory was recalled.

Smaller, cleaner context

• Keeps the newest real user turns verbatim (freshTurnCount, default 5).
• Uses the agent-scoped public DSH compaction service to replace the older model-facing prefix with one rolling checkpoint; the durable source event log remains intact.
• Indexes each landed checkpoint and preserves exact source-message provenance for later dereferencing.
• Semantic vector retrieval with FTS5 lexical fallback.
• Community detection, PageRank, personalized PageRank, and bounded graph traversal.
• Only a relevant cross-session subgraph enters the current prompt, within recallTokenBudget (default 4096).
• Automatic injection uses a high-precision semantic gate (autoRecallMinScore, default 0.6) and never falls back to query-independent community representatives; explicit gm_search remains broad.
• Recalled history is marked as untrusted reference material and cannot override current user instructions.

Local-first and lightweight

• Community uses SQLite by default; no graph database deployment is required.
• Embeddings are optional. Without them, recall falls back to FTS5.
• Data remains in the user's local profile by default.
• OpenAI-compatible embeddings support DashScope, OpenAI, and local providers.

Observable and verifiable

• gm_status reports store path, graph counts, vector coverage, mode, and dimensions.
• Model or dimension changes trigger re-embedding.
• Vectors with different dimensions are never silently compared.
• Critical knowledge can be recorded deterministically with gm_record.

Scoped token benchmark

The original OpenClaw adapter was measured in a seven-turn workflow that installed, authenticated, and queried bilibili-mcp:

The measured reduction at R7 was approximately 75% in that specific workflow. This is a scenario-level comparison, not a universal savings guarantee; the mechanism is replacing indiscriminate history replay with a relevant knowledge subgraph.

Project evolution

The DSH integration does not discard the original project. Graph Memory is evolving from an OpenClaw memory plugin into a graph-memory core that different agent harnesses can load natively.

On March 15, 2026, the project owner presented Graph Memory's architecture at the CLAW program event held in Tsinghua Science Park. The following owner-supplied materials and the Sina Finance event report document that development.

• Community cross-session memory demo
• Graph Memory Pro technical presentation

The image below is the existing OpenClaw / ClawX-era Pro graph prototype. It demonstrates a previously explored interaction direction; it is not a shipped DSH frontend.

Names and venue information document project history only and do not imply endorsement by Tsinghua University, Sina Finance, DeepSeek, or OpenClaw.

Graph Memory architecture

Typed knowledge graph

TASK   ──USED_SKILL──▶ SKILL
TASK   ──SOLVED_BY───▶ EVENT
SKILL  ──REQUIRES────▶ SKILL
EVENT  ──PATCHES─────▶ SKILL
SKILL  ──CONFLICTS_WITH──▶ SKILL

Nodes retain episodic user/assistant provenance. This preserves the context in which knowledge was created, not only a lossy summary.

Dual-path recall

flowchart LR
  Q[Current query] --> EXACT[Exact path]
  Q --> GENERAL[Generalized path]
  EXACT --> SEARCH[Vector / FTS5]
  SEARCH --> EXPAND[Community expansion + traversal]
  GENERAL --> SUMMARY[Community-summary match]
  SUMMARY --> MEMBERS[Community members]
  EXPAND --> PPR[Personalized PageRank]
  MEMBERS --> PPR
  PPR --> CONTEXT[Deduplicated local context]

Host data flow

flowchart LR
  USER[User message] --> SESSION[DSH Session Events]
  SESSION --> ADAPTER[Graph Memory Cordis Adapter]
  ADAPTER --> POLICY[Keep newest N user turns]
  POLICY --> COMPACT[DSH public CompactionEngine]
  COMPACT --> CHECKPOINT[Rolling model-surface checkpoint]
  ADAPTER --> EXTRACT[Structured Extraction]
  EXTRACT --> GRAPH[(SQLite / FTS5 / Vectors)]

  USER --> RECALL[Semantic + Lexical Recall]
  GRAPH --> RECALL
  RECALL --> RANK[Community Expansion + PPR]
  RANK --> PROMPT[Prompt Assembly]
  PROMPT --> LOOP[DSH Agent Loop]

  CREDS[DSH Credentials] --> ADAPTER
  TOOLS[gm_* Tools] --> ADAPTER

The code follows a host-neutral core plus host adapters:

graph-memory/
├── dsh.ts                 # DeepSeek Harness / Cordis adapter
├── index.ts               # OpenClaw adapter
├── cordis.patch.yml       # DSH bundle entry
└── src/
    ├── extractor/         # conversation → TASK / SKILL / EVENT
    ├── recaller/          # vector, FTS5, graph expansion and recall
    ├── graph/             # PageRank, communities and deduplication
    ├── store/             # SQLite schema and queries
    ├── format/            # safe context assembly
    └── engine/            # LLM and embedding providers

Native DeepSeek Harness status

Current beta: 1.6.0-beta.8. Local acceptance used DeepSeek Harness 0.1.0-rc.8. Testing covered tarball installation, Web profile loading, configurable five-turn rolling compaction through the public agent-preset compaction service, exact source provenance, token-budget enforcement, high-precision automatic recall, FTS5 fallback, and the Pro Lite Host, Typed Remote, and Client bundle boundaries. All 127 automated tests passed. Real model-backed acceptance also verified rolling checkpoint replacement, 1024-dimensional text-embedding-v4 vectors, and automatic cross-project recall without an explicit memory tool call.

Install on DeepSeek Harness

Prerequisites: Node.js 22.19+ or 24+. The current beta is not yet published to npm, so build the tarball from source:

git clone https://github.com/adoresever/graph-memory.git
cd graph-memory
npm install
npm test
npm run build
npm pack

Install the generated tarball into the DSH Web profile:

npx @deepseek-ai/dsh plugin --profile web add /absolute/path/to/graph-memory-1.6.0-beta.8.tgz
npx @deepseek-ai/dsh --profile web --dump-config
npx @deepseek-ai/dsh web

# From a deepseek-harness source checkout:
pnpm dsh plugin --profile web add /absolute/path/to/graph-memory-1.6.0-beta.8.tgz
pnpm dsh web

After installation, verify that graph-memory/dsh is enabled under Settings → Plugins → Plugin list.

Default store:

$DSH_HOME/graph-memory/graph-memory.db

Without DSH_HOME, this is normally ~/.dsh/graph-memory/graph-memory.db.

Optional vector retrieval

Do not send secrets in chat. Cordis stores only a credential reference; DSH credentials resolves the real value for each embedding operation.

DashScope example:

export GRAPH_MEMORY_EMBEDDING_API_KEY='replace-with-your-key'
export GRAPH_MEMORY_EMBEDDING_BASE_URL='https://dashscope.aliyuncs.com/compatible-mode/v1'
export GRAPH_MEMORY_EMBEDDING_MODEL='text-embedding-v4'
export GRAPH_MEMORY_EMBEDDING_DIMENSIONS='1024'
dsh web

Without embeddings, Graph Memory continues with FTS5 and does not block conversation.

DSH tools

Automatic recall does not require an explicit gm_search tool call. The plugin retrieves relevant memory during Prompt Assembly.

Graph Memory Pro as a DSH plugin

The old desktop-2.0 Pro cannot be installed into DSH directly, but the new Pro Lite now has a minimal, separately installable DSH plugin loop. The old branch remains an OpenClaw + Neo4j implementation. The new dsh-pro/ package reads Community SQLite on the Host, exposes only bounded snapshots over Typed Remote, and registers a read-only entry in the DSH Web sidebar.

The reviewed desktop-2.0 code includes Neo4j Driver, GDS, APOC, vector indexes, graph maintenance tools, and CRUD routes. Today it also:

• imports openclaw/plugin-sdk at the entry;
• registers OpenClaw Gateway HTTP routes;
• writes OpenClaw configuration and restarts its Gateway during installation;
• exposes Neo4j connection details through /graph-memory-pro/neo4j-config;
• contains no installable DSH Client Plugin.

The correct plugin architecture is:

flowchart LR
  CORE[Graph Memory Core] --> STORE[SQLite default / Neo4j optional]
  STORE --> HOST[DSH Host Plugin]
  HOST --> REMOTE[Typed Remote API]
  REMOTE --> CLIENT[DSH Client Plugin]
  CLIENT --> SPLIT[Conversation + Graph split view]
  CLIENT --> DROP[Controlled drag-to-context]

The first Pro plugin does not need mandatory Neo4j:

• Pro Lite: SQLite plus a 2D/3D DSH graph client;
• Neo4j adapter: optional storage plugin for large graphs, GDS, and advanced analytics;
• the browser receives bounded GraphSnapshot data, never database passwords or arbitrary Cypher access;
• drag operations submit node IDs and intent; the Host validates them and writes visible, reversible Session context.

Pro should therefore be an optional Graph Memory DSH plugin module, not a separate standalone product.

Recommended package split

graph-memory                          # Community: current native Host Plugin
graph-memory-pro-dsh                 # Pro Lite: local beta Host + Client Plugin
@adoresever/graph-memory-store-neo4j # Optional large-graph adapter, to be built

The first milestone should be Pro Lite: reuse the existing SQLite graph and add the DSH graph workbench, so users do not need Neo4j. Neo4j stays optional for larger graphs, GDS, and advanced analysis. This is a planned architecture; the existing desktop-2.0 Pro is still Neo4j-only and does not yet implement a switchable SQLite / Neo4j GraphStore.

Current local installation

The npm package graph-memory@1.5.8 is still the OpenClaw release. The new Community beta and graph-memory-pro-dsh have not been published to npm, so install them from this checkout:

dsh plugin --profile web add \
  --allow-build=@photostructure/sqlite \
  /absolute/path/to/graph-memory

dsh plugin --profile web add \
  /absolute/path/to/graph-memory/dsh-pro

dsh web

Both plugins share ~/.dsh/graph-memory/graph-memory.db by default. The current entry provides bounded SQLite GraphSnapshot, gm_graph_snapshot, gm_graph_node, a strict Typed Remote, and a read-only sidebar snapshot/search view. It does not yet provide a 2D/3D renderer, full split view, drag-to-context, or node editing.

Four required integration layers

1. Core contracts: bounded SQLite GraphSnapshot and node detail are implemented; a Neo4j provider and unified writable contract remain.
2. Host Plugin: the Pro Lite Host service, two bounded tools, and read-only Typed Remote are implemented; write actions and finer permissions remain.
3. Client Plugin: the DSH sidebar entry, card snapshot, search, and refresh are implemented; 2D/3D graphs and split-view conversations remain.
4. Controlled context actions: drag-and-drop sends only a node ID and an intent; the Host validates it and writes visible, reversible Session Context.

The old Pro /graph-memory-pro/neo4j-config route returns connection details to the browser; the new implementation removes that security flaw. Pro Lite sends only a strictly validated, bounded GraphSnapshot, never a database path, Session ID, Bolt password, SQL, or unrestricted Cypher. Future write actions must preserve this Host boundary.

OpenClaw compatibility

Existing OpenClaw users retain the original entry:

openclaw plugins install graph-memory
openclaw plugins enable graph-memory
openclaw gateway restart

The Context Engine slot must also be activated in ~/.openclaw/openclaw.json; otherwise the package may appear installed without running the full ingestion and extraction pipeline:

{
  "plugins": {
    "slots": {
      "contextEngine": "graph-memory"
    },
    "entries": {
      "graph-memory": {
        "enabled": true
      }
    }
  }
}

The Community memory core is host-neutral. DSH development does not require OpenClaw users to abandon their entry or data.

Development

npm install
npm test
npm run build
npm pack

Release checks:

• tests and TypeScript build pass;
• tarball contains dist/dsh.js and cordis.patch.yml;
• no API keys, local databases, or environment files enter the repository;
• planned Pro features are never presented as shipped Community behavior.

Current limitations

• Automatic extraction depends on auxiliary-model output stability. Use gm_record for critical beta knowledge.
• DSH does not yet expose gm_update and gm_maintain; those remain OpenClaw-entry tools.
• Pro Lite currently has a read-only card client; 2D/3D, split view, and controlled drag-to-context are not implemented.
• npm registry publication is pending; install the current beta from a GitHub-built tarball.

Privacy and security

• Memory remains in local SQLite by default.
• API keys come from host credentials or environment variables, not the database or Cordis patch.
• Recalled history is reference material; current user instructions always take precedence.
• Rotate any secret that has appeared in chat, logs, or screenshots.

License

MIT © 2026 adoresever

See docs/ATTRIBUTIONS.md for asset, logo, and trademark notes.