Emre Elbeyoglu

I built markdown.new Put markdown.new before any URL → get clean Markdown back. Cloudflare's Markdown for Agents is great, but only works for enabled sites. markdown.new works for ANY website on the internet. 80% fewer tokens. Also converts PDFs, images, audio. Free. No signup. markdown.new

@timsoulo I actually wrote a tweet about this. You should also optimize your 404 page for visitors who arrive via hallucinated links.

I just discovered a brand new (and totally fascinating) way to come up with content ideas for your blog. It’s your 404 pages. (Stay with me…) I randomly checked if Ahrefs Blog had any 404 pages with lots of backlinks pointing at them. And saw hundreds of them.. Wow! I pinged my Head of Content Ryan Law asking why so much of our blog content was now “dead.” Turns out — none of it was ever alive. These are all hallucinated URLs from AI-generated articles. 🤯 Basically, hundreds of sites have published AI-generated content and never checked whether the links in those articles actually point to real pages. But here's the part that really blew my mind... Most of these hallucinated URLs actually make sense. AI "thinks" we should have articles on: /internal-linking /how-important-are-backlinks /domain-authority /content-decay etc. And... we don't. AI literally found gaps in our content strategy that we missed. So now we're going to publish articles on all these topics and turn hundreds of broken hallucinated links into real backlinks. If you run a blog, go check your 404s right now. You might be sitting on a goldmine of content ideas that AI already "voted" for. (go to Ahrefs' Site Explorer > "Best by links" report > filter by 404 status code) Fun times. 😅

@harimanok_ On file handling, nothing is stored. Files are processed in memory on Cloudflare Workers and discarded immediately after conversion. No persistence, no logging.

@elbeyoglu I’ve been using AWS Textract for similar functionality, and this looks promising. Is it open-source, so we can deploy it to our own Cloudflare account? Also, I didn’t find any information regarding how the files are managed. Are they stored on your server, or are they temporary?

🔥 New on markdown.new: File to Markdown ⚡ Drop any file → get clean, AI-ready Markdown back. 📄 PDFs, DOCX, ODT 📊 Excel, Numbers, CSV 🖼 Images (AI-powered OCR + summarization) Upload or paste a URL. API included, no key needed. Free. No signup. Details: markdown.new/file-to-markdo…

@snowball_cip markdown.new is built for personal use and AI agents. You or your agent paste a URL, get clean Markdown back. Nothing stored, everything processed in memory and discarded immediately.

@elbeyoglu file conversion should be a backend process, not something users need to think about, this just adds to the friction.

@Gtarafdarr I used @Remotion for the animation, and it only took 10 minutes to create. 🔥

@elbeyoglu Thanks for supporting the community. This is cool tool. I love the video animation too. Can you plz share which tool you have used to make this animation?

@MatterTurbulent We are using @CloudflareDev Workers AI OCR

@elbeyoglu was LITERALLY just thinking about needing to make this for myself. ty

@dpshade markdown.new is 8X faster and leaner

@elbeyoglu very cool. how's it compare to github.com/jina-ai/reader…?

@eastdakota One-click deployments for popular open-source projects would be massive 🔥

What should we launch tomorrow? 🤔

@dpshade Coming soon 👀

@elbeyoglu Get YT -> transcript MD well and I'd definitely consider moving some of my products to this

@dpbmaverick98 markdown.new is free and will stay free. But for business users who need higher rate limits, a pay-per-call option like x402 could make a lot of sense. Cool idea!

@elbeyoglu This so needs to be on x402 so people can pay per call and make it self-sustaining. Hey @elbeyoglu, I have a gateway implementation that works on Cloudflare Workers too and can wrap this in a pay-as-you-go endpoint via x402. hmu!

@bygregorr Nothing is stored. Files are processed in memory on Cloudflare Workers and discarded immediately after conversion. No logging, no persistence.

@elbeyoglu What happens when you're working with sensitive data that can't be uploaded freely?

@itsjustmarky Can you give me examples to debug it?

@elbeyoglu Very cool, but it fails on most of the sites I try where r.jina.ai hasn't had any issues.

I built markdown.new Put markdown.new before any URL → get clean Markdown back. Cloudflare's Markdown for Agents is great, but only works for enabled sites. markdown.new works for ANY website on the internet. 80% fewer tokens. Also converts PDFs, images, audio. Free. No signup. markdown.new

Appreciate the perspective! Though I'd push back a bit: I built markdown.new to provide a simple and fast way to get markdown from any URL, that's it. Sometimes you just need clean markdown in one step: no pipeline, no site integration required. You're right about navigation. We're actively working on making the stripping smarter with options to keep or remove nav links depending on the use case. There's already real demand for this. Look at context7.com they serve clean markdown for popular product docs and it's widely used by agents and coding tools. Every token counts. Would love to see what @vercel_dev does in this space 🙏

I don't want to make anyone mad, but both of these approaches are flawed: If agents want to auto-convert your HTML to markdown, they can do so themselves (and popular agents do post-processing before handing content to the main LLM). What does work is to respond with bespoke markdown from the underlying content source that is higher fidelity than what an auto-converted may achieve. Also keep in mind, that you actually don't just want to strip the navigation. Agents need to a way to find other pages as well.

@dann_pvn @CloudflareDev made it fast :) I think your faq part of site is SSR so you need to browser rendering: markdown.new/https://chilli…

@elbeyoglu How can it be so damn fast? I added my website and it almost instantly returned the markdown. One thing it got wrong tho is that it didn't get the answer to the FAQs at the bottom of the page (chillital.com).

Most of them do convert to markdown internally. But caching the whole web has problems: content goes stale, and publishers want control over how AI uses their content. The better path is what's happening now: publishers serve markdown natively via content negotiation. markdown.new just bridges the gap until that's everywhere.

@elbeyoglu why dont the big guys (Anthropic, OpenAI, etc.) do this already? I mean pre-markdown-ify major websites and then get them from cache?

@curtis_sx Can you please share the url you have tried?

@elbeyoglu + @elbeyoglu I tried 30 different url (some works some dont its not the issue) but why after 30 try i get this

bascially will break on any dynamic loaded website or cloudflare protected. This vibecoder just query the server instead of spoofing a browser wait for load then scrapping content. @elbeyoglu wallet and inbox open

@dzimica Yes jina also have the similiar tool but the only thing jina misses is speed. They convert very slow.

@elbeyoglu I get better output with r.jina.ai This only relates to forum threads. Love the tool anyway, I'm going to use it on websites.

Thanks for the feedback! We do support keeping images. Just add retain_images: true to your request. Example (curl): curl markdown.new -H 'Content-Type: application/json' \ -d '{"url": "example.com", "retain_images": true}' Or use the query parameter: markdown.new/example.com?re… By default, images are removed to keep token usage low, but you can include them anytime you need.

@elbeyoglu This is very cool, but also it doesn’t include markdown formatted links to embedded images ie ![]() format. This is mission critical for many applications.

@dipeshdotdev Thank you. The rate limit is 25 requests per day per IP. (for now)

@elbeyoglu Simple yet useful tool. Nice work men. Btw are there any rate limits or something like that? Like what if someone ran a script with 10k unique urls and used your service? Have you planned for that?

@gm_mertd Not really. X and LinkedIn block datacenter IPs, so the requests get rejected before we can fetch the page. This is a limitation of any server-side scraping tool, not just markdown.new. Sites that require login or block bots won't work.

@elbeyoglu does it work with x or Linkedin?