MailWise — Your Team's Email Memory
Stop re-investigating issues your team already solved. MailWise turns your email threads into a searchable knowledge base — index thousands of EML files locally, tag expert engineers, and use RAG to instantly surface how similar issues were debugged and resolved in the past.
"Have we seen this before?" → MailWise searches 25,000+ emails in under a second and shows you exactly what your best engineers said about it.
What you get
Semantic search — not keyword matching, but meaning-based retrieval across your entire email archive:
CODEBLOCK0
RAG-powered analysis — Claude reads the most relevant past threads and synthesizes debugging guidance:
CODEBLOCK1
Get started in 3 steps
CODEBLOCK2
That's it. Now search with mailwise search "..." or get AI analysis with mailwise analyze "...".
Commands
Search for similar issues
CODEBLOCK3
- -
--show-body — show a preview of each matching message - INLINECODE3 — only show replies from expert engineers
- INLINECODE4 — number of results (default: 10)
Deep analysis with RAG
CODEBLOCK4
- -
-k N — number of similar issues to feed to Claude (default: 5)
Tip: Paste the FULL bug report — error codes, logs, environment details produce much better matches.
CODEBLOCK5
View a full email thread
CODEBLOCK6
Check index status
CODEBLOCK7
Manage expert engineers
CODEBLOCK8
Expert engineers' replies get ★Expert tags and boosted scores in search results.
Typical workflow
- 1. User describes an issue or pastes a bug report
- Run
mailwise search "..." --show-body to find similar past issues - If promising results found, run
mailwise analyze "..." for deep RAG analysis - Use
mailwise show <ID> to read full threads of interest - Summarize findings: root cause patterns, debugging steps, and next actions
Privacy & data handling
- - Local-only commands (
index, search, show, stats, experts): Everything runs on your machine. Embeddings generated locally via sentence-transformers. No data sent anywhere. - External LLM command (
analyze): Sends email excerpts to the Anthropic API via Claude Code CLI. Do not use on sensitive emails unless your org's data policy permits it.
Authentication
- - The
analyze command requires Claude Code CLI — run claude once to authenticate, or set INLINECODE18 - No auth needed for local commands (
index, search, show, stats, experts)
MailWise — 你的团队邮件记忆库
别再重复调查团队已解决的问题。 MailWise将你的邮件线程转化为可搜索的知识库——在本地索引数千个EML文件,标记专家工程师,并使用RAG技术即时呈现过去类似问题的调试与解决过程。
我们之前遇到过这个问题吗? → MailWise在不到一秒内搜索25,000+封邮件,精准展示你的顶尖工程师对此问题的见解。
你能获得什么
语义搜索 — 不是关键词匹配,而是基于语义的检索,覆盖整个邮件存档:
$ mailwise search 文件夹迁移后日历同步失败 --show-body
#1 [0.89] ★专家 RE: 邮箱迁移后Outlook日历无法同步
来自: senior.engineer@example.com (2024-11-15)
根本原因是过时的文件夹ID缓存。清除HxCalendarSync
并重启同步服务...
#2 [0.84] RE: 迁移后同步失败 — Mac客户端
来自: support.lead@example.com (2024-10-22)
我们之前见过这种模式。检查迁移工具是否
保留了文件夹GUID...
#3 [0.81] ★专家 RE: 移动后日历项目消失
来自: senior.engineer@example.com (2024-09-08)
这与CASE-4521属于同一类bug。修复方法是...
RAG驱动分析 — Claude读取最相关的历史线程并综合调试指导:
$ mailwise analyze 用户报告移动到新文件夹后日历事件消失
基于5个类似历史问题,你的团队发现:
● 根本原因模式:迁移后文件夹ID缓存过期(5例中的3例)
● 推荐第一步:检查HxCalendarSync日志中的GUID不匹配
● 专家共识:清除同步缓存可解决约80%的此类问题
● 边缘情况:如果迁移过程中未保留GUID,则需要完全重新同步
三步上手
bash
pip install mailwise # 1. 从PyPI安装
mailwise init # 2. 交互式设置向导 — 创建配置,设置目录
mailwise index # 3. 索引你的EML文件(增量式,可安全重复运行)
就这样。现在使用 mailwise search ... 搜索,或使用 mailwise analyze ... 获取AI分析。
命令
搜索类似问题
bash
mailwise search 在此描述问题 --show-body
- - --show-body — 显示每条匹配邮件的预览
- --expert-only — 仅显示专家工程师的回复
- -k N — 结果数量(默认:10)
使用RAG进行深度分析
bash
mailwise analyze 在此粘贴完整bug报告或问题描述
- - -k N — 提供给Claude的类似问题数量(默认:5)
提示: 粘贴完整的bug报告 — 错误代码、日志、环境细节能产生更好的匹配结果。
bash
mailwise analyze $(cat bug_report.txt) -k 10
查看完整邮件线程
bash
mailwise show <邮件ID>
检查索引状态
bash
mailwise stats
管理专家工程师
bash
mailwise experts list
mailwise experts add engineer@company.com --name 张三
mailwise experts remove engineer@company.com
专家工程师的回复会获得 ★专家 标签,并在搜索结果中获得加权评分。
典型工作流程
- 1. 用户描述问题或粘贴bug报告
- 运行 mailwise search ... --show-body 查找类似历史问题
- 如果找到有希望的结果,运行 mailwise analyze ... 进行深度RAG分析
- 使用 mailwise show 阅读感兴趣的完整线程
- 总结发现:根本原因模式、调试步骤和后续行动
隐私与数据处理
- - 本地命令(index、search、show、stats、experts):一切在你的机器上运行。通过sentence-transformers在本地生成嵌入向量。不发送任何数据。
- 外部LLM命令(analyze):通过Claude Code CLI向Anthropic API发送邮件摘录。除非你的组织数据政策允许,否则不要在敏感邮件上使用。
身份验证
- - analyze 命令需要 Claude Code CLI — 运行 claude 一次进行身份验证,或设置 ANTHROPICAPI_KEY
- 本地命令(index、search、show、stats、experts)无需身份验证