Vesper

About

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

• Three-layer memory architecture: Working, Semantic, and Procedural
• HippoRAG for knowledge graph-driven multi-hop reasoning
• Voyager-style procedural memory for learning user skills and workflows
• Simple local setup with Docker and pre-configured MCP tools
• 1 GitHub stars
• High-performance semantic search using BGE-large embeddings

Use Cases

• Improving AI agent answer quality and contextual understanding over time
• Enabling AI agents to remember and adapt to user preferences and workflows across conversations
• Enhancing AI agent capabilities with personalized and learning memory