Ollama + OpenClaw Local Memory & Vector Search System for Openclaw

First, fetch the portable release of Ollama. We recommend downloading the ZIP directly or using a git-mirror:

# Download the portable ZIP into your temp folder
Invoke-WebRequest -Uri "https://www.modelscope.cn/models/ollama/ollama-windows/resolve/master/ollama-windows-amd64.zip" -OutFile "D:/temp/ollama-windows-amd64.zip"

Next, extract the binaries into the designated OpenClaw environment directories:

# Create the local directory structure
New-Item -ItemType Directory -Force -Path "$env:APPDATA/mx/openclaw-home/$env:USERNAME/.openclaw/tools/ollama"
New-Item -ItemType Directory -Force -Path "$env:APPDATA/mx/openclaw-home/$env:USERNAME/.openclaw/ollama-data/models"

# Extract the portable files
Expand-Archive -Path "D:/temp/ollama-windows-amd64.zip" -DestinationPath "$env:APPDATA/mx/openclaw-home/$env:USERNAME/.openclaw/tools/ollama" -Force

Configure your local models environment and run the background service:

$env:OLLAMA_MODELS = "$env:APPDATA/mx/openclaw-home/$env:USERNAME/.openclaw/ollama-data/models"
Set-Location "$env:APPDATA/mx/openclaw-home/$env:USERNAME/.openclaw/tools/ollama"
Start-Process -FilePath ".\\ollama.exe" -ArgumentList "serve" -WindowStyle Hidden

Pull the required vector embedding model:

.\\ollama.exe pull nomic-embed-text

Now, update your .openclaw/openclaw.json configuration file to point to your local Ollama instance and enable memory-tencentdb:

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "enabled": true,
        "provider": "openai",
        "model": "nomic-embed-text",
        "remote": {
          "baseUrl": "http://localhost:11434/v1",
          "apiKey": "ollama"
        }
      }
    }
  },
  "plugins": {
    "enabled": true,
    "allow": ["mx", "memory-tencentdb"],
    "load": {
      "paths": ["D:/Users/YOUR_USERNAME/AppData/Roaming/mx/openclaw-home/YOUR_USERNAME/.openclaw/plugins/memory-tencentdb"]
    },
    "slots": {
      "memory": "memory-tencentdb"
    }
  }
}

Finally, rebuild your vector index for existing archives:

pip install requests tqdm
python scripts/rebuild_vector_index.py --incremental

The local memory and vector search system organizes data in a clean directory hierarchy, caching vector embeddings alongside raw databases for optimized processing: