Zaxy

•
17
•

Provides an event-sourced temporal knowledge graph fabric for AI agent memory, replacing markdown files and vector databases with a structured, replayable bi-temporal system.

Zaxy serves as an advanced memory system for AI agents, offering a structured, replayable, and bi-temporal knowledge graph that surpasses traditional markdown files and vector databases. It is built on Eventloom for an immutable, append-only JSONL audit trail with SHA-256 hash chains, and leverages Neo4j for its temporal graph capabilities, allowing facts to have validity windows. Zaxy features sophisticated hybrid extraction and retrieval mechanisms, combining rule-based processing with LLM fallbacks and fusing exact, keyword, vector, and graph traversal queries, making it an ideal choice for developing intelligent agents that require durable, auditable, and context-rich memory capabilities.

主な機能

01Immutable audit trail using Eventloom with SHA-256 hash chains.
02Bi-temporal knowledge graph where facts include validity windows (valid_from, valid_to).
03Hybrid extraction engine combining rule-based processing with LLM fallback for efficiency.
04Hybrid retrieval across exact matches, keywords, vector embeddings, and graph traversal.
05MCP-native integration for seamless memory management within any MCP-compatible agent framework.
061 GitHub stars

ユースケース

01Developing AI agents that require structured, auditable, and temporal memory capabilities.
02Replacing traditional unstructured memory solutions like markdown files and vector databases for agent memory.
03Integrating advanced memory management into MCP-compatible agent frameworks such as LangGraph.