Git 历史比你的站立会议了解得更多。我们开发了一款 AI 来查询它。
我已经花了一段时间在构建 Gitmore。我要解决的问题是:每周我都要花 1-2 小时浏览提交记录,试图拼凑出“我们发布了什么?”这个问题的答案,而利益相关者并不想阅读“feat: 实现 oauth2 及刷新”的信息。
我的洞察是:所有信息都已经在 Git 中。提交、拉取请求、作者、时间戳。对于非开发人员来说,这些信息几乎是不可读的。
因此,我在其上构建了一个 AI 层,具体功能包括:
1. 通过 OAuth 连接到 GitHub/GitLab/Bitbucket(仅读取元数据,从不读取代码)
2. 通过 Webhook 实时捕获提交和拉取请求
3. 使用 Claude 将原始 Git 活动转化为人类可读的摘要
4. 根据您希望的时间表自动通过电子邮件或 Slack 发送
示例转化:
```
转换前:“fix: 删除过时的 API 调用,refactor: 提取认证中间件”
转换后:“通过更新过时的端点修复了 API 超时问题。通过集中认证逻辑提高了安全性。”
```
技术栈:Next.js 15、MongoDB、Bull 队列用于异步报告生成,Claude API 用于摘要。使用 Webhook 实时获取数据,而不是轮询。
在构建这个过程中我学到的一些事情:
- 提交信息遵循严格的模式(73% 以 feat:/fix:/refactor: 开头),但几乎没有“为什么”的上下文
- 团队每年每人花费约 78 小时手动编写状态报告
- “我们这周发布了什么?”这个问题占据了关于代码库查询的 62%
从同一数据层中出现的其他功能:
- 可以聊天的 AI 代理(“Sarah 上周做了什么?”)
- 带有贡献评分的开发者排行榜
- 自动生成的公共变更日志
免费套餐:1 个代码库,1 个自动化。专业版($15/月):5 个代码库。企业版($49/月):20 个代码库 + 自定义品牌报告。
https://gitmore.io
欢迎提问有关架构、AI 提示策略或 Webhook 处理的技术问题。我也很好奇,其他团队是如何处理“我们发布了什么”这个问题的?
查看原文
I've been building Gitmore for a while now. The problem I was solving: every week
I'd spend 1-2 hours scrolling through commits trying to piece together "what did
we ship?" for stakeholders who don't want to read "feat: impl oauth2 w/ refresh".<p>The insight: all the information is already in Git. Commits, PRs, authors, timestamps.
It's just unreadable for anyone who isn't a developer.<p>So I built an AI layer on top that:<p>1. Connects to GitHub/GitLab/Bitbucket via OAuth (reads only metadata, never code)
2. Captures commits and PRs in real-time via webhooks
3. Uses Claude to transform raw Git activity into human-readable summaries
4. Delivers automatically via email or Slack on whatever schedule you want<p>Example transformation:<p><pre><code> Before: "fix: rm deprecated api calls, refactor: extract auth middleware"
After: "Fixed API timeouts by updating deprecated endpoints. Improved
security by centralizing authentication logic."
</code></pre>
Technical stack: Next.js 15, MongoDB, Bull queues for async report generation,
Claude API for summarization. Webhooks for real-time data, not polling.<p>Some things I learned building this:<p>- Commit messages follow strict patterns (73% start with feat:/fix:/refactor:)
but contain almost no "why" context
- Teams spend ~78 hours/year/person writing status reports manually
- The question "what did we ship this week?" accounts for 62% of queries
about repositories<p>Other features that emerged from the same data layer:
- AI agents you can chat with ("What did Sarah work on last week?")
- Developer leaderboards with contribution scoring
- Auto-generated public changelogs<p>Free tier: 1 repo, 1 automation. Pro ($15/mo): 5 repos.
Enterprise ($49/mo): 20 repos + custom branded reports.<p>https://gitmore.io<p>Happy to answer technical questions about the architecture, AI prompting strategy,
or webhook handling. Also curious - how do other teams handle the "what did we ship"
problem?