持久化 AI 聊天会话：跨刷新与关闭继续运行，免去超时和状态管理

来源证据

用 request/response 端点构建聊天代理会遇到超时和回合间无记忆，开发者通常要自行加入 Postgres、Redis 和队列协调，才能让流在刷新后存活。

memory between turns. So you write everything to Postgres, add Redis so the stream survives a refresh, and push the slow work onto a queue that you then have to coordinate. That's a lot of plumbing before your agent does anything interesting. chat.agent gives every conversation its own machine instead. It lives for the whole conversation, sleeps when nobody's typing, and wakes up where it left off. What that gets you: No timeouts. In our production data 1 in 20 turns runs longer than 36 minutes,