Show HN: Rubi – A knowledge graph that cuts how much Claude Code re-reads (github.com)

Rubi, a new knowledge graph and context manager for AI coding agents, has been developed to enhance the efficiency of models like Claude Code by significantly reducing the costs associated with re-reading conversation history. In extensive testing of 284 Claude sessions, it was discovered that a staggering 71% of costs stemmed from context re-reading, particularly in sessions extending beyond 150,000 tokens. Rubi addresses these issues by maintaining shorter sessions while ensuring continuity through smarter context compaction and organizing project knowledge effectively through a temporal knowledge graph. This system organizes decisions, bugs, and concepts, allowing AI models to quickly access relevant information without sifting through extensive backlogs of data. This advancement is significant for the AI/ML community as it demonstrates the potential for reducing operational costs linked to AI coding sessions while optimizing the way these models process and retrieve information. Rubi employs a combination of techniques from information retrieval and graph theory, including Personalized PageRank for contextual relevance and BM25 for efficient search capabilities. These improvements have led to reported reductions in session costs by approximately 37% during testing and the promise of even larger savings as its use becomes widespread. By enabling more effective context management, Rubi not only reduces costs but can enhance the overall productivity of AI coding assistants, paving the way for more sophisticated and resource-efficient AI applications.