Vanguard Memory Node (VMN)

Local deterministic memory for AI agents via the Model Context Protocol (MCP).

No cloud. No vector database. No semantic drift. Your data stays on your NVMe.

How it works

VMN uses xLMP (Logical Memory Protocol) — a deterministic, content-addressed memory substrate. It shatters text into fixed shards and retrieves evidence using lexical keyword density scoring, not probabilistic cosine similarity.

Install

npm install -g vanguard-memory-node

Claude Desktop Integration

Add to your claude_desktop_config.json:

Mac

~/Library/Application Support/Claude/claude_desktop_config.json

Windows

%APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "vanguard-memory": {
      "command": "node",
      "args": ["/path/to/vanguard_memory_node/dist/index.js"]
    }
  }
}

For WSL on Windows:

{
  "mcpServers": {
    "vanguard-memory": {
      "command": "wsl",
      "args": ["-d", "Ubuntu", "node", "/home/edt/vanguard_memory_node/dist/index.js"]
    }
  }
}

Tools

vmn_ingest

Stores text locally as a SHA-256 addressed shard.

• Input: text (string)
• Returns: SHA-256 hash

vmn_recall

Retrieves evidence from a shard using deterministic xLMP lexical search.

• Input: hash (string), query (string)
• Returns: bounded evidence window

Local vault

~/.vanguard/local_vault/<hash>.txt

Built by ExergyNet

https://exergynet.org