Beads upgrades AI agent memory with graph-based tracking

Steve Yegge’s "memory upgrade" addresses the fundamental "amnesia" problem in agentic development by treating task state as structured, queryable data. It uses Dolt for Git-like branching and merging of the agent's memory alongside the codebase, while machine-optimized JSON output reduces token overhead and improves LLM parsing reliability. As a core component of the "Gas Town" ecosystem for coordinating autonomous agent swarms, Beads employs graph-based dependency tracking to prevent agents from getting lost in complex refactors. Its open-source and local-first architecture provides a privacy-conscious alternative to cloud-based agent platforms.