Yesterday’s surprise leak of the source code of Anthropic’s cloud code revealed a lot about the vibe-coding scaffolding the company has built around its proprietary cloud model. But observers also discovered references to disabled, hidden or inactive features, searching more than 512,000 lines of code in more than 2,000 files that provide a glimpse of a possible roadmap for future features.
Chief among these features is Kairos, a persistent daemon that can run in the background even when the Cloud Code terminal window is closed. The system will use periodic “
Kairos uses a file-based “memory system” designed to allow persistent operations across user sessions. A hint hidden behind a disabled “KAIROS” flag in the code explains that the system is designed to “have a complete picture of who the user is, how they want to collaborate with you, what behaviors to avoid or repeat, and the context behind the work the user gives you.”
To organize and consolidate this memory system across sessions, the cloud source code includes references to an explicitly named AutoDream system. When a user becomes inactive or manually tells Anthropic to sleep at the end of a session, the AutoDream system will tell the cloud code that “you are completing a dream – a mirrored pass over your memory files.”
This prompt describing this AI “dream” process tells the cloud code to scan the day’s transcripts for “new information worth retaining”, consolidate that new information in a way that avoids “near-duplicates” and “contradictions”, and prune existing memories that are overly verbose or have recently become outdated. The cloud code will also be instructed to keep an eye on “existing memories that have overflowed”, a problem we’ve seen before when cloud users have attempted to graft memory systems onto their harnesses. According to the prompt, the overall goal will be to “synthesize what you have recently learned into durable, well-organized memories so that future sessions can be quickly oriented.”
<a href