Gomaa

Gomaa: The Local-First Memory Operating System for AI Agents

Gomaa is an advanced AI memory platform designed to address common challenges faced by autonomous agents. It offers a persistent memory system that integrates seamlessly with Obsidian vaults, facilitating powerful semantic search capabilities, graph memory management, and robust security features. Key aspects of Mnemosyne include:

Key Features

• Hierarchical Scoping: Organize memories into clearly-defined categories known as Wings (projects/domains) and Rooms (topics), ensuring that unrelated data does not contaminate search results.
• Verbatim Ingestion: Capture and index conversation transcripts with 100% accuracy, preserving the integrity of coding snippets, API keys, and specific syntax.
• Ebbinghaus Temporal Decay: Optimize memory retention by allowing frequently accessed memories to remain at the forefront while gradually decaying unused memories over time, enhancing organization.
• Memory Version Management: Track changes through memory versioning, enabling rollback and inspection of previous versions for complete history oversight.
• Zero External API Cost: Operates entirely on local resources, running embeddings on CPU/GPU without incurring additional API costs, making it a cost-effective solution.
• Obsidian Markdown Vault Integration: Store all memory data in human-readable Markdown format, allowing for easy access and manipulation through Obsidian.
• MCP Compliance: Compatible with the Model Context Protocol (MCP), Mnemosyne facilitates integration with various tools in the AI ecosystem.

The Purpose of Mnemosyne

Traditional memory systems can lead to:

• Context Window Overload: Systems that create large operational bottlenecks by allowing chat histories to expand excessively.
• Context Contamination: Mixing unrelated domain data in vector databases, leading to inefficient searches.
• Lossy Summaries: Tools that summarize past interactions may lose important details critical for AI reasoning.
• Memory Degradation: Irrelevant, outdated memories making searches less effective due to lack of adaptive decay.

By addressing these challenges, Gomaa delivers a production-ready solution suitable for operational AI agents, enhancing their ability to remember, retrieve, and reason within their contexts efficiently.

How It Works

Gomaa employs a sophisticated system architecture combining hybrid graph and semantic vector retrieval. This enables agents to manage interaction and memory in a streamlined, effective manner. Using a clear hierarchy allows for efficient memory searches and retrievals, while an embedded Ebbinghaus decay engine ensures memories are properly prioritized based on usage.

Example Usage

from mnemosyne.core import UnifiedMemorySystem

# Initialize the memory engine
memory = UnifiedMemorySystem()

# Storing a memory with hierarchical context
memory.remember(
    title="PostgreSQL Production Cluster Setup",
    content="Primary cluster operates at 172.16.8.2:5432 with pgvector 0.7 enabled.",
    tags=["infra", "database"],
    wing="devops",
    room="databases",
    salience=0.9
)

# Scoped hybrid memory retrieval
results = memory.recall(
    query="Where is the postgres cluster running?",
    mode="hybrid",  # Merges semantic similarity + keyword + graph links
    scope={"wing": "devops"}
)

print(results[0]["title"], "->", results[0]["content"])

Mnemosyne provides an effective platform for managing AI memory, taking the complexity out of remembering and retrieving critical information. With comprehensive features and an efficient operational model, it is well-suited for deployment in diverse AI-driven applications.