Background

A surprise leak of the source code for Anthropic’s Claude Code revealed a substantial codebase—over 512,000 lines across more than 2,000 files. The dump gave observers a rare look at the internal scaffolding that supports the Claude model, including sections that are currently disabled or hidden.

Hidden Daemon: Kairos

Among the most notable discoveries is a component named Kairos. The code describes Kairos as a persistent daemon capable of operating even when the Claude Code terminal window is closed. It is set to issue periodic “” prompts that assess whether new actions are required. A special “PROACTIVE” flag is also defined, intended to “surface something the user hasn’t asked for and needs to see now.”

A disabled flag in the code includes a prompt explaining Kairos’s purpose: to develop “a complete picture of who the user is, how they’d like to collaborate with you, what behaviors to avoid or repeat, and the context behind the work the user gives you.” This suggests a vision of an AI that maintains a continuous understanding of user preferences and context.

Memory Management: AutoDream

The source also references an evocatively named AutoDream system. According to the embedded prompts, when a user goes idle or manually instructs Anthropic to “sleep” at the end of a session, AutoDream would initiate a “dream”—a reflective pass over the AI’s memory files. The dream process is described as scanning the day’s transcripts for “new information worth persisting,” consolidating that information while avoiding “near‑duplicates” and “contradictions,” and pruning existing memories that are overly verbose or outdated.

AutoDream is further instructed to watch for “existing memories that drifted,” a problem previously observed when users attempted to graft memory systems onto their AI harnesses. The overall goal, as stated in the code, is to “synthesize what you’ve learned recently into durable, well‑organized memories so that future sessions can orient quickly.”

Implications

These hidden components indicate that Anthropic is exploring mechanisms for continuous‑state operation and automated knowledge management. Kairos could enable the model to act proactively, delivering insights without explicit prompts, while AutoDream aims to keep the AI’s internal knowledge base clean, relevant, and readily accessible across sessions. Together, they point toward a future where large language models maintain an evolving memory of user interactions, potentially improving personalization and efficiency.

While the features are currently disabled, their presence in the codebase suggests they are under active development or testing. Observers will likely watch how Anthropic integrates—or chooses not to integrate—these capabilities into upcoming releases of Claude.

Used: News Factory APP - news discovery and automation - ChatGPT for Business