Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

How a RAG Agent Became a Personal Secretary

AI summary

The author transformed a standard RAG Agent that routed questions to specialist agents and retrieved from ChromaDB into a personal secretary with task memory, scheduled reminders, daily briefings, and WeCom integration. SQLite provides transactional storage for concurrent CLI, Streamlit, and daemon access, while conditional updates make reminder delivery idempotent and avoid duplicate notifications. The author used planning, execution, and review agents for development but retained architectural decisions and key constraints as human responsibilities.

In-depth insight

The report offers a concrete implementation pattern for turning a retrieval chatbot into a persistent, proactively scheduled assistant, while candidly identifying limitations in conversation continuity, intent reasoning, task prioritization, and decomposition.

起点

我的个人秘书系统原本是个标准 RAG Agent:Supervisor 路由给学业/生活两个专家,Worker 检索本地 ChromaDB,MemoryCurator 每轮写长期记忆。本质是一个更聪明的搜索引擎——问,答,结束。

我要的是主动和持久:记住要做的事、盯着时间、到点提醒、每天早上主动汇报。从"答疑机器人"变成"有办公桌的助理"。

一、架构决策:先定死,再动手

1. 存储:SQLite,理由是并发正确性 这个系统至少三个进程并发读写同一份数据——CLI、Streamlit、提醒 daemon。JSON 是"全量读→改→全量写",两进程并发改就互相覆盖;要救就得自己加文件锁,等于造一个没有事务的数据库。

sqlite3 是标准库,WAL 下"一写多读"并发良好。

#ChromaDB#SQLite#Streamlit#WeCom#DeepSeek V4 Pro#DeepSeek V4.1 Flash#GLM

Original · 9 min readRead the original
Group chat

No comments yet. Start the conversation.