@omega-memory/openclaw

OMEGA persistent memory plugin for OpenClaw — graph-based, local-first, #1 on LongMemEval benchmark (95.4%).

Replaces OpenClaw's built-in memory with OMEGA's semantic graph memory. Your agent remembers conversations, learns preferences, and builds context across sessions — all running locally, no subscription required.

Install

openclaw plugins install @omega-memory/openclaw

Requires OMEGA (pip install omega-memory or uvx omega-memory).

What it does

Feature	Description
Auto-recall	Before each AI turn, relevant memories are injected as context
Auto-capture	After each turn, important information is extracted and stored
Graph memory	Related memories are linked, boosting recall beyond keyword match
User profile	Persistent profile that builds over time
Lessons	Cross-session learning from verified patterns

Agent tools

The plugin exposes four tools to the LLM:

• memory_recall — semantic search across all stored memories
• memory_store — explicitly save a fact, decision, preference, or lesson
• memory_forget — delete a specific memory by ID
• memory_profile — read or update the persistent user profile

Slash commands

• /memories — show recent memories from the past 7 days
• /remember <text> — quick-save something to memory
• /omega-status — OMEGA health metrics and memory statistics

Configuration

In your OpenClaw config, set plugins.slots.memory to "omega-memory":

{
  "plugins": {
    "slots": {
      "memory": "omega-memory"
    },
    "omega-memory": {
      "transport": "stdio",
      "serverCommand": "uvx omega-memory",
      "autoRecall": true,
      "autoCapture": true,
      "maxRecallResults": 8,
      "project": "openclaw"
    }
  }
}

Option	Default	Description
transport	"stdio"	"stdio" spawns a local server, "sse" connects to a running one
serverCommand	"uvx omega-memory"	Command to start OMEGA (stdio mode)
serverUrl	—	Server URL (SSE mode only)
autoRecall	true	Inject relevant memories before each AI turn
autoCapture	true	Store important info after each AI turn
maxRecallResults	8	Max memories injected per turn
profileFrequency	25	Inject user profile every N turns
project	"openclaw"	OMEGA project scope for memory isolation

How it works

User message → [before_agent_start hook]
                    ↓
              omega_query(message) → ranked memories
                    ↓
              inject <omega-memory> context block
                    ↓
              LLM generates response
                    ↓
              [agent_end hook]
                    ↓
              omega_store(user message) → persisted

The plugin is a thin TypeScript bridge over the MCP protocol. All intelligence — embedding, graph linking, deduplication, consolidation — runs in the OMEGA server process.

vs alternatives

	Built-in	LanceDB	Supermemory	OMEGA
Local-first	Yes	Yes	No	Yes
Semantic search	No	Yes	Yes	Yes
Graph relationships	No	No	No	Yes
User profile	No	No	Yes	Yes
Free	Yes	Yes	No	Yes
Benchmarked	No	No	No	#1 LME
Needs API key	No	OpenAI	Supermemory	No

License

MIT