aura-turbo

Compressed Persistent Memory for AI Agents

Give your AI assistant a memory that never forgets — and fits in your pocket.

aura-turbo is an MCP (Model Context Protocol) server that provides persistent, compressed memory storage for Claude and other AI assistants. Memories survive across sessions, are compressed with zlib for efficient storage, and are searchable by keyword and tag.

Features

• Persistent memory — SQLite-backed storage that survives restarts
• zlib compression — Typical 3-5x compression on text content
• Three memory levels — core (permanent), working (7-day TTL), ephemeral (24h TTL)
• Tag-based organization — Categorize and retrieve memories by tag
• Keyword search — TF-IDF-style relevance scoring across all memories
• Automatic expiry — Ephemeral and working memories clean up after themselves
• Deduplication — Content-hash prevents storing identical memories twice
• Zero external dependencies — Uses only Python stdlib (zlib, sqlite3) plus the MCP SDK

Installation

pip install aura-turbo

Or clone and run directly:

git clone https://github.com/aura-turbo/aura-turbo.git
cd aura-turbo
pip install -e .

Quick Start

Run the server directly:

python server.py

Claude Desktop Configuration

Add to your Claude Desktop MCP config (claude_desktop_config.json):

{
  "mcpServers": {
    "aura-turbo": {
      "command": "python",
      "args": ["/path/to/aura-turbo/server.py"]
    }
  }
}

Or if installed via pip:

{
  "mcpServers": {
    "aura-turbo": {
      "command": "aura-turbo"
    }
  }
}

Custom Database Location

Set the AURA_TURBO_DB environment variable to use a custom database path:

{
  "mcpServers": {
    "aura-turbo": {
      "command": "python",
      "args": ["/path/to/aura-turbo/server.py"],
      "env": {
        "AURA_TURBO_DB": "/path/to/my-memories.db"
      }
    }
  }
}

Default location: ~/.aura-turbo/memories.db

Tools

memory_store

Store a memory with optional tags and importance level.

content: "The user prefers dark mode and uses vim keybindings"
tags: ["preferences", "editor"]
level: "core"

memory_recall

Search memories by keyword or phrase. Returns ranked results.

query: "editor preferences"
limit: 5

memory_forget

Delete a specific memory by ID.

memory_id: 42

memory_stats

Show storage statistics — total memories, compression ratio, space saved.

memory_search_by_tag

Find all memories with a specific tag.

tag: "preferences"

memory_clear_expired

Remove ephemeral memories older than 24h and working memories older than 7 days. Core memories are never removed.

How Compression Works

aura-turbo uses Python's built-in zlib library (which implements DEFLATE/gzip compression) to compress memory content before storing it in SQLite.

Content Type	Typical Ratio
English prose	3-4x
Code snippets	3-5x
JSON/structured data	4-6x
Already compressed	~1x

A 1KB conversation summary compresses to ~250-300 bytes. Over hundreds of memories, this adds up to significant space savings while keeping everything in a single portable SQLite file.

Comparison

Feature	aura-turbo	Mem0	Custom RAG
Price	Free	$249/mo	Varies
Storage	Local SQLite	Cloud	Cloud/Local
Dependencies	MCP SDK only	Many	Many
Compression	zlib (3-5x)	None	None
Setup time	1 minute	Account setup	Hours
Privacy	100% local	Cloud	Depends
MCP native	Yes	No	No

Architecture

Claude/AI <--MCP--> server.py <--> storage.py <--> SQLite + zlib
                                        |
                                   ~/.aura-turbo/
                                    memories.db

License

MIT