read-anything · 读不完的收藏，让 AI 替你读懂

Too much saved, too little read — let your AI read it for you.

English · 中文

演示：丢一个链接进去 → 自动抓取 / 转写 → 出中文结构化解读（精读稿 / 快扫 / 访谈拆解…）→ 就着材料继续追问。 A link in → auto-fetch / transcribe → a structured Chinese explainer → keep asking follow-ups over the source.

你：https://www.xiaoyuzhoufm.com/episode/xxxx 这期讲了啥
它：（本地转写音频 + shownotes → 访谈拆解：两位嘉宾各自观点、分歧点、金句带时间戳）
你：他们在 Agent 记忆那段具体怎么说的？
它：（只基于转写回答，标时间戳；没提到的如实说没提到）

English

Is your bookmarks folder full of things you "plan to get to someday" — tens of thousands of words of long-form deep dives, an hour-long English podcast, a forty-minute Bilibili video, dozens of X threads… that just gather dust because you can't finish them, or can't even get through them?

read-anything helps you understand them fast: drop in a link, and a few minutes later you get a Chinese explainer that spells out "what this actually says," and you can keep asking follow-up questions about it. It all happens inside your AI agent, the results land as markdown, and they accumulate into your own knowledge base.

The specific hassle it solves

• Saved it but no time to read → in a few minutes you get an explainer that "makes sense even without the original"
• It's English / video / audio, and a pain to get through → auto-transcription + a structured Chinese explainer straight out
• Want to dig into details → converse over the material; answers are based only on the source, never made up
• Read it and forgot it, scattered everywhere → explainers land as markdown, accumulating into your own system
• Don't want to bounce between a dozen web tools → one link, all inside your agent

Formats it can handle

Web pages / blogs / single WeChat public-account articles / YouTube / Bilibili / Xiaoyuzhou podcasts / X posts / local recordings (meeting recordings, voice memos — just drop the file path for m4a/mp3/wav) / plain text (including pasted meeting notes, speech-to-text transcripts).

English content produces a Chinese explainer directly (no need to translate first and then read); videos, podcasts, and local recordings without captions go through local transcription (faster-whisper — audio never leaves your machine, zero API cost).

Document links behind a login wall like Feishu/Notion can't be fetched (it will tell you so honestly) — just export the text transcript from Miaoji and paste it in, and it will break it down all the same.

Six reading perspectives (+ your own)

你想要	用哪个模板
完整读懂，不看原文也行	精读稿（默认）
30 秒判断值不值得细读	快扫
播客 / 对谈，按人拆观点	访谈拆解
会议录音/纪要，拆决策·待办·分歧	会议纪要
arXiv / 技术报告	论文拆解
从内容里挖可写的选题和素材	选题素材

Not enough? Drop one of your own .md files into templates/ and it becomes a new template — just say "use my xx template."

One principle: if it can't get it, it says so

Fetch failures, only getting a summary, audio that didn't transcribe — it tells you all of it honestly, and never forces out a polished-looking explainer from incomplete material. Transcription output is labeled "may contain transcription errors," and quoted highlights remind you to check against the original. Every sentence you read is traceable.

Install

Claude Code (one command):

git clone https://github.com/heyuxuan0209/read-anything.git ~/.claude/skills/read-anything