展示HN:我们分析了1,573个Claude Code会话,以了解AI代理的工作方式。
Show HN: Rudel – Claude Code Session Analytics

原始链接: https://github.com/obsessiondb/rudel

## Rudel:Claude 代码分析工具 Rudel 提供了一个仪表盘,用于分析你的 Claude 代码编程会话,提供关于 token 使用量、会话时长、模型性能等方面的洞察。它可以通过 CLI 工具轻松集成:使用 `npm install -g rudel` 安装,登录,并使用 `rudel enable` 启用自动上传。 Rudel 捕获详细的会话数据——记录、时间戳、git 上下文和子代理使用情况——并将其安全地存储在 ClickHouse 中进行分析。可以通过仪表盘授予团队访问权限。 过去的会话也可以批量上传。 **重要安全提示:** Rudel 摄取*完整*的会话数据,可能包括敏感信息,例如代码和密钥。仅在上传此数据可接受的项目中使用它。 虽然 Rudel 不会通过托管服务访问记录中的个人数据,但在使用前请查看他们的隐私政策。 在他们的 GitHub 仓库中可以找到 CLI 命令、自托管和安全报告的文档。 在 [app.rudel.ai](https://app.rudel.ai) 免费试用托管版本。

## Rudel.ai:分析Claude Code会话 一个团队构建了[rudel.ai](https://rudel.ai)来深入了解他们自己对Claude Code的使用情况,由此产生了一个包含1,573个会话(1500万+ tokens,27万+交互)的数据集,他们现在将其开源作为分析工具。 主要发现包括技能使用率出乎意料地低(4%的会话),26%的会话放弃率(通常在第一分钟内),以及根据任务而异的成功率——文档编写最高,重构最低。他们还识别出在最初两分钟内出现的“错误级联”模式,该模式强烈预测会话放弃。 开发者们正在构建一个“良好”代理性能的基准,承认目前缺乏标准。另一位开发者分享了他们互补的工作,即使用阶段和自动化门来构建会话,从而将接受率从73%提高到90%以上。 Rudel.ai 免费使用,并提供云端和本地(Docker)部署选项。该团队欢迎提问和贡献,并致力于改进对其他模型(如Codex)的分析提取。
相关文章

原文

Try the hosted version for free at rudel.ai

Analytics for Claude Code. Rudel gives you a dashboard with insights on your coding sessions — token usage, session duration, activity patterns, model usage, and more.

  1. Create an account at app.rudel.ai
  2. Install the CLI and connect it to your account:
npm install -g rudel

rudel login     # authenticate via your browser
rudel enable    # auto-upload sessions when Claude Code exits
  1. Invite teammates (optional): go to Settings → Organization in the dashboard, enter their email, and share the generated invite link with them.

That's it. Your Claude Code sessions will now be uploaded automatically.

Already have past sessions? Upload them in one go:

rudel upload    # interactive picker for batch upload

See the CLI documentation for all available commands.

  1. You install the CLI and run rudel enable
  2. This registers a Claude Code hook that runs when a session ends
  3. The hook uploads the session transcript to Rudel
  4. Transcripts are stored in ClickHouse and processed into analytics

Each uploaded session includes:

  • Session ID & timestamps (start, last interaction)
  • User ID & organization ID
  • Project path & package name
  • Git context (repository, branch, SHA, remote)
  • Session transcript (full prompt & response content)
  • Sub-agent usage

Security & Privacy Disclaimer

Rudel is designed to ingest full coding-agent session data for analytics. That means uploaded transcripts and related metadata may contain sensitive material, including source code, prompts, tool output, file contents, command output, URLs, and secrets that appeared during a session.

Only enable Rudel on projects and environments where you are comfortable uploading that data. If you use the hosted service at app.rudel.ai, we do not have access to personal data contained in uploaded transcripts and cannot read that data. Review the Rudel Privacy Policy before enabling uploads for yourself or your team.

See CONTRIBUTING.md for local setup, development commands, and PR guidelines.

For self-hosting your own instance, see docs/self-hosting.md.

To report a vulnerability, see SECURITY.md. Do not open public issues for security concerns.

MIT

联系我们 contact @ memedata.com