AI 产品案例 · Agent

AI 每日早报

一个 Hermes Agent cron job，每天 9 点读互联网，整理成有个性的 AI 早报推送到飞书。

运行中 · 已连续推送

RESOURCES

GitHub：https://github.com/Techdoll00/morning-brief
真实样本：https://github.com/Techdoll00/morning-brief/tree/main/samples

01问题

刷推特 2 小时/天才能跟住 AI 圈动态；newsletter 是 link dump 没人帮你连回自己；推给我的和我无关。

02方案

Hermes Agent 每早 9 点触发：抓 GitHub Trending（筛 AI/Agent）+ Tavily 搜索 + Exa 检索 → DeepSeek V4 Pro 整理 + 连点 + 写人话 → 飞书 DM 推送。

03我的职责

Prompt 设计（含人格）、cron 配置、飞书集成链路、错误兜底。

01 · 背景与痛点

离开 Style3D 实习后意识到：资讯不是不够，而是没有人为"我"过滤和连点。

02 · 产品设计与 Agent 架构

方案：Hermes Agent 每早 9 点触发：抓 GitHub Trending（筛 AI/Agent）+ Tavily 搜索 + Exa 检索 → DeepSeek V4 Pro 整理 + 连点 + 写人话 → 飞书 DM 推送。

我的职责：Prompt 设计（含人格）、cron 配置、飞书集成链路、错误兜底。

工作流设计

graph LR
A[9:00 CST cron] --> B[GitHub Trending]
A --> C[Tavily Search]
A --> D[Exa Search]
B & C & D --> E[DeepSeek V4 Pro 整理]
E --> F[飞书 DM]

03 · 复盘

每天 9 点推送的稳定性比内容更重要：抓取失败、模型超时、推送失败各做了一层 fallback，才敢挂 cron 连续跑。复盘：cron 类产品的信任来自‘连续 N 天没断过’，而不是某一天多精彩。

04 · 成果与产品思考

2026 年 7 月起每天 9 点准时推送；samples/ 里是真实输出，不是 mockup。

产品思考：证明你能做 AI 产品最好的方式，是先做一个给自己用的。

AI product case · Agent

AI Daily Brief

A Hermes Agent cron job that reads the internet at 9 AM and pushes a curated, personable AI briefing to Feishu.

Running · Daily push

RESOURCES

GitHub: https://github.com/Techdoll00/morning-brief
真实样本: https://github.com/Techdoll00/morning-brief/tree/main/samples

01Problem

2 hours/day on Twitter just to stay current; newsletters are link dumps that don't connect back to your situation; what's pushed to me isn't relevant.

02Approach

Hermes Agent cron at 9 AM daily: GitHub Trending (filtered for AI/Agent) + Tavily search + Exa retrieval → DeepSeek V4 Pro curates, connects dots, writes warmly → Feishu DM push.

03My role

Prompt design (with persona), cron configuration, Feishu integration chain, error handling.

01 · Background & Pain

After leaving the Style3D internship, I realized: information is not scarce — nobody filters and connects it for me.

02 · Product Design & Agent Architecture

Approach: Hermes Agent cron at 9 AM daily: GitHub Trending (filtered for AI/Agent) + Tavily search + Exa retrieval → DeepSeek V4 Pro curates, connects dots, writes warmly → Feishu DM push.

My role: Prompt design (with persona), cron configuration, Feishu integration chain, error handling.

Workflow design

graph LR
A[9:00 CST cron] --> B[GitHub Trending]
A --> C[Tavily Search]
A --> D[Exa Search]
B & C & D --> E[DeepSeek V4 Pro 整理]
E --> F[飞书 DM]

03 · Pitfalls & Hard Lessons

[TBD: share 1-2 of the hardest debugging / cost-cutting / architecture decisions]

04 · Outcome & Product Thinking

Daily 9 AM push since July 2026; samples/ contains real outputs, not mockups.

Product thinking: The best way to prove you can build AI products is to build one that serves you first.

TopicsAgent · Product · Build