Git 历史比你的站会更懂你。我们构建了一个 AI 来查询它。
1 分•作者: slmslm•6 天前
我一直在开发 Gitmore 一段时间了。我试图解决的问题是：每周我都会花 1-2 个小时浏览提交记录，试图拼凑出“我们发布了什么？”给那些不想阅读“feat: impl oauth2 w/ refresh”的利益相关者看。<p>我的见解是：所有信息都已存在于 Git 中。提交记录、PR、作者、时间戳。只是对于非开发人员来说，这些信息难以理解。<p>所以我构建了一个 AI 层，它：<p>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 实时捕获提交记录和 PR
3. 使用 Claude 将原始 Git 活动转换为人类可读的摘要
4. 按照您想要的任何时间表，通过电子邮件或 Slack 自动发送<p>转换示例：<p><pre><code> 之前：“fix: rm deprecated api calls, refactor: extract auth middleware”
之后：“通过更新已弃用的端点修复了 API 超时问题。通过集中身份验证逻辑提高了安全性。”
</code></pre>
技术栈：Next.js 15、MongoDB、用于异步报告生成的 Bull 队列、用于摘要的 Claude API。使用 Webhook 进行实时数据传输，而不是轮询。<p>我从中学习到的一些东西：<p>- 提交消息遵循严格的模式（73% 以 feat:/fix:/refactor: 开头），但几乎不包含“为什么”的上下文
- 团队每年/每人花费大约 78 小时手动编写状态报告
- “我们这周发布了什么？”这个问题占关于存储库查询的 62%<p>从相同的数据层中涌现的其他功能：
- 您可以与之聊天的 AI 代理（“Sarah 上周做了什么？”）
- 具有贡献评分的开发者排行榜
- 自动生成的公共变更日志<p>免费套餐：1 个存储库，1 个自动化。Pro（每月 15 美元）：5 个存储库。企业版（每月 49 美元）：20 个存储库 + 自定义品牌报告。<p>https://gitmore.io<p>很乐意回答有关架构、AI 提示策略或 Webhook 处理的技术问题。也很好奇——其他团队如何处理“我们发布了什么”这个问题？
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?