LiteMemory

Overview

LiteMemory is the persistent memory layer for the Lite Suite, now absorbed into LiteCore. At its core is LCM (Lossless Context Management) — a DAG-based memory system that stores every message immutably and automatically compacts context at increasing depth intervals (minutes, hours, days, months). This means your AI agents never lose context, even across long-running sessions, without blowing up token budgets.

Obsidian vault access is included as a secondary feature for reading and writing notes, but the primary architecture is the LCM memory graph.

Features

• LCM (Lossless Context Management) — DAG-based memory architecture that preserves full conversation history as an immutable graph
• Automatic compaction at increasing depth intervals — raw messages compact into summaries at minute, hour, day, and month boundaries
• Immutable message storage — nothing is ever deleted, only summarized at higher levels
• FTS5 full-text search across all stored memory
• Absorbed into LiteCore — runs as a built-in module, not a standalone service
• Obsidian vault read/write access by path or title
• Vault full-text search, folder listing, and tag-based discovery
• Daily note access — read today's note or append entries
• Link traversal — follow wikilinks between notes
• RAG indexing and semantic search via vault_rag and vault_rag_index

Installation

LiteMemory is installed automatically by the Lite Suite installer. To install manually:

cd C:/Apps/LiteMemory
pnpm install
pnpm build

The installer registers the MCP server with Claude Desktop and any other configured MCP clients.

Usage

Once registered, LiteMemory tools are available in any MCP-compatible AI session. Example interactions:

• "Search my vault for notes on project planning" → uses vault_search
• "Add a log entry to today's daily note" → uses vault_daily
• "Read my note on API design patterns" → uses vault_read
• "What notes are tagged #architecture?" → uses vault_tags

Configuration

LiteMemory requires one environment variable:

| Variable | Description | |----------|-------------| | OBSIDIAN_VAULT_PATH | Absolute path to your Obsidian vault directory |

Set this in your environment or in the MCP client config block.

Claude Desktop (claude_desktop_config.json)

{
  "mcpServers": {
    "litememory": {
      "command": "node",
      "args": ["C:/Apps/LiteMemory/dist/index.js"],
      "env": {
        "OBSIDIAN_VAULT_PATH": "C:/Users/YourName/Documents/Obsidian Vault"
      }
    }
  }
}

Available MCP Tools

| Tool | Description | |------|-------------| | vault_read | Read a note by path or title | | vault_write | Create or overwrite a note | | vault_search | Full-text search across all notes | | vault_list | List notes in a folder | | vault_daily | Read or append to today's daily note | | vault_recent | List recently modified notes | | vault_tags | Find notes by tag | | vault_link | Follow or discover wikilinks from a note | | vault_rag | Semantic search using the RAG index | | vault_rag_index | Build or refresh the RAG index |

Requirements

• Node.js 18 or later
• An Obsidian vault on the local filesystem
• OBSIDIAN_VAULT_PATH must point to the vault root (the folder containing .obsidian/)
• For RAG features: sufficient RAM to hold the index in memory (scales with vault size)

Troubleshooting

"Vault not found" error on startup Verify that OBSIDIAN_VAULT_PATH points to the correct directory and that the path contains a .obsidian folder. On Windows, use forward slashes or escaped backslashes in the config.

vault_search returns no results The search index is built on first run. If the vault is large, allow a few seconds on startup. If results are still missing, restart the MCP server to trigger a re-index.

vault_rag returns an error about missing index Run vault_rag_index first to build the semantic index before using vault_rag. The RAG index is not built automatically to avoid slow startup on large vaults.