Claude Persistent Memory

About

This system provides Claude Code sessions with long-term memory that persists across conversations, transforming it into an AI assistant that truly remembers. It combines hybrid BM25 full-text and vector semantic search for powerful retrieval, structures memories automatically via an LLM into a consistent XML format, and leverages 4-channel retrieval via MCP tools and various hooks. Memories are automatically clustered, gain confidence through validation feedback, and can be promoted to reusable skills, ensuring your AI assistant constantly learns and retains critical information.

Key Features

• LLM Structuring: Automatically structures memories into a standardized XML format using an LLM.
• Hybrid Search: Combines BM25 full-text and vector semantic similarity for enhanced retrieval.
• 4-Channel Retrieval: Supports on-demand memory pulling via MCP tools and automatic injection via hooks.
• Automatic Clustering: Groups similar memories, promoting mature clusters to reusable skills.
• Confidence Scoring & Time Decay: Memories gain/lose confidence through feedback and fade over configurable half-lives.
• 0 GitHub stars

Use Cases

• Providing Claude Code with long-term memory that persists across development sessions.
• Automating the extraction and structuring of knowledge from Claude Code conversations.
• Maintaining stable facts, architectural decisions, bug fixes, and user preferences within Claude Code.