Memory Vector

Name: Memory Vector
Rating: 3.5 (1 reviews)
Author: media-builders

By media-builders 👁 65 views ▲ 0 votes

OpenClaw plugin for semantic memory search over local markdown memory using LanceDB

GitHub

Install

npm install
python3

Configuration Example

{
  "memoryPaths": [
    "<WORKSPACE_ROOT>/agents/<AGENT_OR_WORKSPACE>/memory",
    "<WORKSPACE_ROOT>/agents/<AGENT_OR_WORKSPACE>/MEMORY.md"
  ]
}

README

# Memory Vector

`memory-vector` is an OpenClaw plugin that adds semantic search over local memory markdown files using Sentence Transformers and LanceDB.

## Tools

The plugin registers four OpenClaw tools:

- `memory_vector_ingest`: discovers memory files, chunks them, generates embeddings, and rebuilds the LanceDB table
- `memory_vector_search`: searches indexed memory with a natural-language query
- `memory_vector_watch`: starts or stops background indexing for changed memory files
- `memory_vector_status`: reports the current vector index state

## What It Ingests

The portable install path is `memoryPaths`: exact memory files or folders chosen during installation.

```json
{
  "memoryPaths": [
    "<WORKSPACE_ROOT>/agents/<AGENT_OR_WORKSPACE>/memory",
    "<WORKSPACE_ROOT>/agents/<AGENT_OR_WORKSPACE>/MEMORY.md"
  ]
}
```

When a folder is listed in `memoryPaths`, markdown files under that folder are ingested. When a single `.md` file is listed, only that file is ingested.

Git commit history is collected from every git repository discovered under `workspaceRoot`, regardless of which `memoryPaths` are configured.

## Documentation

- [Installation guide](docs/INSTALLATION.md)
- [AI installation instructions](docs/AI_Installation_Instructions.md)

## Quick Install Shape

Add the plugin to `~/.openclaw/openclaw.json`:

```json
{
  "plugins": {
    "entries": {
      "memory-vector": {
        "enabled": true,
        "config": {
          "workspaceRoot": "<WORKSPACE_ROOT>",
          "memoryPaths": [
            "<MEMORY_PATH>"
          ],
          "indexPath": "plugins/memory-vector/vector",
          "embeddingModel": "all-MiniLM-L6-v2",
          "maxSearchResults": 20,
          "pythonPath": "python3",
          "watchEnabled": false,
          "watchDebounceSeconds": 2
        }
      }
    },
    "allow": [
      "memory-vector"
    ],
    "load": {
      "paths": [
        "<PLUGIN_DIR>"
      ]
    }
  }
}
```

Then install dependencies from the plugin directory:

```bash
npm install
python3 -m venv scripts/.venv
scripts/.venv/bin/pip install -r scripts/requirements.txt
```

Restart the Gateway after config changes:

```bash
openclaw gateway restart
```

Build the index with:

```text
memory_vector_ingest
```

tools