Can I self-host Memory Crystal?

Yes, Memory Crystal offers flexible deployment options. You can use its hosted cloud service, self-host it using your own Convex deployment for full data sovereignty, or run it locally with SQLite for context engine functionality only.

What is Memory Crystal?

Memory Crystal is a persistent cognitive memory layer designed for AI agents. It enables them to capture conversations, extract crucial insights, and recall relevant information, fostering continuous learning and informed responses.

What are the key memory features offered by Memory Crystal?

Memory Crystal boasts a dual-layer memory system (Short-term for verbatim messages, Long-term for extracted facts), an intelligent Knowledge Graph to connect related memories, five distinct memory stores (sensory, episodic, semantic, procedural, prospective), and adaptive recall modes.

Which AI platforms and tools is Memory Crystal compatible with?

Memory Crystal ships as an OpenClaw plugin and an MCP server, making it compatible with any MCP-compatible host like Claude Desktop, Cursor, and Windsurf. It also provides a Next.js dashboard for management.

How does Memory Crystal improve AI agent performance?

It significantly enhances AI agents by employing a 'Context Engine' that injects relevant short-term and long-term memories before each AI response. This ensures agents don't forget past interactions and can provide consistent, fully informed answers.

Memory Crystal: AI Agents Remember with Persistent Cognitive Memory

Memory Crystal serves as a sophisticated persistent cognitive memory layer designed for AI assistants, ensuring they remember and learn across interactions. It diligently captures every conversation, intelligently extracts crucial information, and stores it within a sophisticated vector-indexed knowledge graph. This active memory system, dubbed the Context Engine, operates before each AI response, injecting the most relevant short-term and long-term memories to inform the AI's answers. By continuously feeding past conversations into future responses and dynamically connecting related concepts within a knowledge graph, Memory Crystal empowers AI agents to overcome the limitations of forgetting, providing them with a deeper understanding of ongoing context and historical knowledge.

주요 기능

012 GitHub stars

02Two Memory Layers: Short-term memory (STM) for verbatim recent messages and Long-term memory (LTM) for extracted facts, decisions, and lessons, both searched simultaneously.

03Context Engine: An active memory system that injects relevant short-term and long-term memories before each AI response.

04Knowledge Graph: Asynchronously connects related memories (decisions, people, rules) to enhance contextual recall and understanding of relationships.

05Five Memory Stores: Organizes memories into `sensory`, `episodic`, `semantic`, `procedural`, and `prospective` categories with distinct retention and search weights.

06Adaptive Recall: Automatically selects from General, Focused, or Deep search modes based on the context of the user's query.

사용 사례

01Empowering AI agents to retain conversational context and learned knowledge across sessions for continuous, informed interactions.

02Integrating advanced persistent memory capabilities into AI frameworks like OpenClaw or any MCP-compatible host.

03Developing AI assistants that leverage a vector-indexed knowledge graph to understand relationships and provide nuanced, context-aware responses.