oG-Memory Context Engine - OpenClaw Plugin

OpenClaw context-engine plugin backed by oG-Memory. It connects OpenClaw lifecycle calls to the oG-Memory HTTP API, injects layered memory context before model inference, persists new memories after each turn, and supports oG-Memory-owned compact takeover.

Prerequisites

• Node.js >= 18
• OpenClaw installed and running
• An oG-Memory HTTP service reachable from the OpenClaw gateway, or local-mode prerequisites for starting AGFS + ContextEngine from this plugin

Install

cd openclaw_context_engine_plugin
openclaw plugins install -l .

Configure

Set the context engine slot in ~/.openclaw/openclaw.json.

Remote Mode

Use remote when oG-Memory is already running.

{
  "plugins": {
    "slots": {
      "contextEngine": "og-memory-context-engine"
    },
    "entries": {
      "og-memory-context-engine": {
        "enabled": true,
        "config": {
          "mode": "remote",
          "memoryApiBaseUrl": "http://127.0.0.1:8090",
          "prefetchEnabled": true,
          "prefetchTopK": 5,
          "compactTakeoverEnabled": true,
          "summaryMaxChars": 4000,
          "shortTermIndexMode": "sync"
        }
      }
    }
  }
}

Local Mode

Use local when the OpenClaw gateway should start AGFS and ContextEngine.

{
  "plugins": {
    "slots": {
      "contextEngine": "og-memory-context-engine"
    },
    "entries": {
      "og-memory-context-engine": {
        "enabled": true,
        "config": {
          "mode": "local",
          "projectRoot": "/path/to/oG-Memory",
          "agfsServerDir": "/path/to/oG-Memory/agfs",
          "agfsDataDir": "/path/to/ogmem-data",
          "contextEnginePort": 8090,
          "llmProvider": "openai",
          "llmApiKey": "sk-...",
          "llmBaseUrl": "https://api.openai.com/v1",
          "llmModel": "gpt-4o-mini"
        }
      }
    }
  }
}

Restart OpenClaw after changing the config.

openclaw restart

How It Works

OpenClaw gateway
  |
  | lifecycle call
  v
openclaw_context_engine_plugin/index.js
  |
  | HTTP
  v
oG-Memory /api/v1/{prefetch,compose,after_turn,prepare_compaction,compact,dispose}

Main lifecycle behavior:

OpenClaw method	oG-Memory calls	Behavior
assemble()	optional /prefetch, then /compose	Stages quick retrieval hits, then returns layered messages for Profile, Archive History, Session State, and Working Set.
afterTurn()	/after_turn	Sends completed turn messages for incremental extraction and persistence.
compact()	/prepare_compaction, then /compact when takeover is enabled	Runs pre-compaction extraction, validates a one-time prepare token, commits/archive synchronously, and returns a compact summary.
dispose()	/dispose	Releases session resources and lets oG-Memory persist session state.

compactTakeoverEnabled=false keeps OpenClaw responsible for final compact generation, but the plugin still calls prepare_compaction so oG-Memory can extract the delta before OpenClaw compacts the local transcript.

Project Structure

openclaw_context_engine_plugin/
├── index.js                 # Plugin entry, config resolver, HTTP bridge, lifecycle handlers
├── compaction-boundary.js   # Persists OpenClaw compaction boundary entries during takeover
├── process.js               # Local-mode health checks, port probing, process helpers
├── compact.test.js          # Node tests for config, prefetch, and compact takeover behavior
├── openclaw.plugin.json     # Plugin manifest and config schema
├── package.json
├── ENV.md                   # Environment variable reference
└── README.md

Verify

openclaw doctor
node openclaw_context_engine_plugin/compact.test.js

For a remote service, also verify oG-Memory directly:

curl http://127.0.0.1:8090/api/v1/health

Gateway logs should show [og-memory] lines for compose, after_turn, and compact control flow.

Switching Back

To revert to the built-in context engine:

{
  "plugins": {
    "slots": {
      "contextEngine": "legacy"
    }
  }
}

Then restart the gateway.