OpenMemory is an innovative framework designed to provide AI agents with real long-term memory, moving beyond traditional methods like vector searches or retrieval-augmented generation (RAG). With a focus on self-hosting, local-first functionality, and explainable interactions, OpenMemory allows AI systems to manage memory efficiently and effectively.

Key Features

• Cognitive Memory Engine: Unlike traditional vector databases, OpenMemory offers a complete cognitive memory engine. It facilitates the addition of memory to AI agents in a single line of code.

• Local and Scalable: Operates locally with zero cloud dependency, ensuring data ownership and privacy. This setup is especially beneficial for environments with strict regulatory frameworks.

• Multi-sector Cognitive Structure: Allows AI systems to manage various types of memories (e.g., semantic, episodic) seamlessly, enhancing their ability to reason and interact.

• Temporal Knowledge Graph: Provides time-aware memory capabilities, allowing agents to track historical changes in data and manage facts dynamically, enabling accurate reasoning and planning.

Sample Code Usage

Integrating OpenMemory into an AI application is straightforward. Here’s a simple example demonstrating its functionality:

from openmemory import OpenMemory

# Creating an instance of OpenMemory
om = OpenMemory(mode="local", path="./memory.db", tier="deep")
# Adding a memory entry
om.add("User allergic to peanuts", userId="user123")
# Querying the memory
results = om.query("allergies", filters={"user_id": "user123"})
# Returns: [{"content": "User allergic to peanuts", "score": 0.89}]

Standalone Mode

OpenMemory can operate without a backend server, making it simpler than ever to integrate into Node.js or Python applications. This mode results in zero configuration requirements and makes local storage of data straightforward and privacy-centric.

Migration and Compatibility

OpenMemory includes robust tools for migrating memories from popular systems such as Mem0, Zep, and Supermemory. The framework supports various ingestion formats, enabling it to adapt to existing datasets with ease.

Benchmark Performance

• Achieves an impressive average response time of 115 ms for recall operations with 95% accuracy across 100,000 nodes.
• 338 QPS throughput demonstrates its efficiency in high-demand scenarios.

Security and Privacy

OpenMemory employs AES-GCM encryption, ensuring that all user data remains secure and private. It promotes user isolation and offers configurable telemetry options.

For more detailed documentation, visit the OpenMemory GitHub Repository.

Explore the power of memory within AI systems and unlock greater potential for personalized and intelligent interactions with OpenMemory.