AI 编程助手最大的痛点之一是"失忆"——换个工具、换个会话就从零开始。我做了一套本地记忆系统：Claude Code、Codex、Cursor 共用同一个记忆池，每次工作自动存交接、下次自动召回，跨工具无缝接力。One of the biggest pains with AI coding assistants is amnesia — switch tools or sessions and you start from scratch. I built a local memory system: Claude Code, Codex and Cursor share one memory pool, auto-saving a handoff each session and recalling it the next, seamlessly across tools.
本地 BM25 + 向量检索，不配置第三方 LLM、几乎零成本；记忆是数据不是指令，带来源标注和信任边界。它既是我每天在用的基础设施，也开放给学员创业社群会员——和网上那种单机记忆模板完全是两回事。Local BM25 + vector retrieval, no third-party LLM, near-zero cost; memory is treated as data, not instructions, with source labels and trust boundaries. It’s infrastructure I use daily, and it’s open to community members — a world apart from the single-machine memory templates floating around online.