hearth

Every AI you talk to forgets you. Hearth makes them remember.

Hearth is a local-first persistent AI memory system. Install it once, and every AI tool you use — Claude Desktop, LM Studio, Cursor — can store and recall memories across conversations. Your data stays on your machine in a single SQLite file.

git clone https://github.com/reallyreallyryan/hearth.git
cd hearth
pip install -e .
hearth init

That's it. You now have a working memory system. See the Install Guide for detailed step-by-step instructions.

1. Find your hearth path:

   which hearth

1. Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

   {
     "mcpServers": {
       "hearth": {
         "command": "/full/path/to/hearth",
         "args": ["serve"]
       }
     }
   }

Replace /full/path/to/hearth with the output from which hearth.

1. Quit and reopen Claude Desktop (Cmd+Q, then relaunch).

hearth init prints the exact config with the full path filled in — just copy and paste it.

Add to ~/.lmstudio/mcp.json:

{
  "mcpServers": {
    "hearth": {
      "command": "/full/path/to/hearth",
      "args": ["serve"]
    }
  }
}

hearth init                    # Set up ~/hearth/ — database, config, pull embedding model
hearth init --no-models        # Set up without Ollama (cloud-only, keyword search only)
hearth serve                   # Start the MCP server (stdio transport)
hearth status                  # Show memory count, embedding status, Ollama availability
hearth remember "some fact"    # Store a memory from the command line
hearth search "query"          # Search your memories

Options for remember: -c category (general, learning, pattern, reference, decision), -p project, -t "tag1,tag2"

Options for search: -p project, -c category, -n limit

When