Getting Started

Install and configure OpenClaw Cortex in about 5 minutes

OpenClaw Cortex is a hybrid semantic memory system for AI agents. It stores memories as both structured metadata and high-dimensional vectors in Memgraph, then retrieves them using a multi-factor scoring algorithm that combines semantic similarity, recency, frequency, type priority, project scope, and graph-traversal signals.

Prerequisites#

Before you begin, make sure you have:

• Docker (for Memgraph)
• Ollama installed and running
• An Anthropic API key (for capture; not required for store / recall / search)

Step 1: Install the binary#

Install via the install script:

curl -fsSL https://raw.githubusercontent.com/ajitpratap0/openclaw-cortex/main/scripts/install.sh | bash

Verify the installation:

openclaw-cortex --version

Or build from source:

git clone https://github.com/ajitpratap0/openclaw-cortex
cd openclaw-cortex
go build -o bin/openclaw-cortex ./cmd/openclaw-cortex
export PATH="$PWD/bin:$PATH"

Step 2: Start Memgraph#

Use the provided docker-compose.yml:

docker compose up -d

Memgraph will be available at:

• Bolt: bolt://localhost:7687 (used by openclaw-cortex)
• HTTP Lab UI: http://localhost:7444

Step 3: Pull the embedding model#

ollama pull nomic-embed-text

Step 4: Store your first memory#

openclaw-cortex store "Always run tests before merging to main" \
  --type rule \
  --scope permanent \
  --tags ci,testing

Step 5: Recall memories#

openclaw-cortex recall "What are the testing requirements?"

You should see the memory from Step 4 returned with a relevance score.

Step 6: Capture from a conversation#

For automatic memory extraction from conversations, set your API key:

export ANTHROPIC_API_KEY=sk-ant-...

Then capture a conversation turn:

openclaw-cortex capture \
  --user "How should I handle errors in Go?" \
  --assistant "Always return errors explicitly. Use fmt.Errorf with %w to wrap them for unwrapping. Never use panic for expected error conditions."

This sends the conversation turn to Claude Haiku, which extracts structured memories, named entities, and relationship facts, then stores them in Memgraph automatically.

Step 7: Wire up Claude Code hooks#

To get automatic memory injection in every Claude Code conversation, add the hook configuration to .claude/settings.json in your project:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "openclaw-cortex hook pre"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "openclaw-cortex hook post"
          }
        ]
      }
    ]
  }
}

Or use the installer:

openclaw-cortex hook install --project my-project

Verify everything works#

# Health check (verifies Memgraph, Ollama, Claude LLM)
openclaw-cortex health

# Check stats
openclaw-cortex stats

# Search memories
openclaw-cortex search "error handling"

Configuration#

The default configuration works out of the box if Memgraph and Ollama are running locally. To customize, create ~/.openclaw-cortex/config.yaml:

memgraph:
  uri: bolt://localhost:7687
  username: ""
  password: ""

ollama:
  base_url: http://localhost:11434
  model: nomic-embed-text

memory:
  dedup_threshold: 0.92
  default_ttl_hours: 720

Next Steps#

• Read the Configuration reference
• Explore the CLI reference
• Learn about the Architecture
• Set up Claude Code Hooks for automatic memory