@cortex-ai-memory/cortex-memory

Persistent memory plugin for OpenClaw — powered by Cortex.

4-tier memory architecture (Working → Episodic → Semantic → Procedural) with Bayesian beliefs, people graph, and multi-signal retrieval. Local-first, zero-cloud, Rust-native.

Prerequisites

Install the Cortex MCP server binary:

# From source
git clone https://github.com/gambletan/cortex.git
cd cortex && cargo install --path cortex-mcp-server

# Or copy the binary to PATH
cp target/release/cortex-mcp-server ~/.local/bin/

Install

openclaw plugin add @cortex-ai-memory/cortex-memory

Configuration

In your OpenClaw config:

{
  "plugins": {
    "@cortex-ai-memory/cortex-memory": {
      "dbPath": "~/.cortex/memory.db",
      "binaryPath": "cortex-mcp-server",
      "autoCapture": true,
      "autoRecall": true,
      "topK": 10
    }
  }
}

| Option | Default | Description | |--------|---------|-------------| | dbPath | ~/.cortex/memory.db | Path to SQLite database | | binaryPath | auto-detect | Path to cortex-mcp-server binary | | autoCapture | true | Store conversation context after each turn | | autoRecall | true | Inject relevant memories before each turn | | topK | 10 | Max memories to retrieve |

Tools

The plugin registers these tools for your AI agent:

• memory_search — Search memories by query (ranked by similarity, recency, salience)
• memory_store — Save information to persistent memory
• memory_get — Get comprehensive context from all memory tiers
• belief_observe — Update beliefs with Bayesian evidence
• fact_add — Store structured facts (subject-predicate-object triples)
• preference_set — Store user preferences
• person_resolve — Cross-channel identity resolution

CLI

openclaw cortex search "user preferences"
openclaw cortex context
openclaw cortex beliefs
openclaw cortex store "User prefers dark mode"

License

MIT