Provides an intelligent memory system for AI agents with semantic search, knowledge graphs, and multi-hop reasoning.
Vesper is an intelligent, local memory system designed for AI agents, specifically for Claude Code, that prioritizes learning over mere recall. It addresses the limitations of standard conversational history and basic key-value memory by implementing a sophisticated three-layer architecture: Working Memory for recent interactions, Semantic Memory powered by HippoRAG for deep knowledge graph traversal, and Procedural Memory for learning skills and workflows. This innovative approach ensures AI agents not only remember facts but also learn user preferences, contextual patterns, and adapt over time, significantly improving answer quality and personalization with ultra-low latency.
Key Features
01Three-layer memory architecture: Working, Semantic, and Procedural
02HippoRAG for knowledge graph-driven multi-hop reasoning
03Voyager-style procedural memory for learning user skills and workflows
04Simple local setup with Docker and pre-configured MCP tools
051 GitHub stars
06High-performance semantic search using BGE-large embeddings
Use Cases
01Improving AI agent answer quality and contextual understanding over time
02Enabling AI agents to remember and adapt to user preferences and workflows across conversations
03Enhancing AI agent capabilities with personalized and learning memory
