Yesterday’s A surprise source code leak for Anthropic’s Claude Code The vibe-coding that the company built around the proprietary Claude model revealed a lot about the scaffolding. But the observers, who examined more than 512,000 lines of code in more than 2,000 files, also found references to disabled, hidden, or inactive features that hint at a potential roadmap for future features.
Among these features, it occupies the main place KairosClaude Code is a persistent daemon that can run in the background even when the terminal window is closed. The system will use it from time to time
Kairos uses a file-based “memory system” designed to allow continuous operation between user sessions. Demand Hiding behind the inactive “KAIROS” flag in the code, the system explains that the system is designed to “get a complete picture of who the user is, how they want to engage with you, what behaviors to avoid or repeat, and the context behind what the user is giving you.”
Includes references to the Claude Code source code to organize and consolidate this memory system across sessions the affectionately named AutoDream system. When the user is idle or manually orders Anthropic to sleep AutoDream system at the end of the session Say Claude Code that “you are carrying out a dream – a reflective passage over your memory files.”
Describing this artificial intelligence “sleeping” process, the instruction asks Claude Code to scan the day’s transcripts for “new information worth pursuing,” combine that new information in a way that avoids “near duplicates” and “contradictions,” and prune existing memories that are too detailed or newly emerged. Claude Code will also be tasked with taking care of “drifting existing memories”, a problem we’ve seen before. When Claude users try to move their storage systems to their mounts. According to the requirements, the overall goal will be to “synthesize recent learning into continuous, well-organized memories for rapid orientation of future sessions.”
