PDFs suck. If you want to build standout RAG apps, you need tooling that is tailored to your needs. You should own your doc parsing infra. Today we’re launching on @ycombinator .

Join our Discord and become a part of the Chunkr community

Want to customize it? Check out the complete open-source repo at

Use your own keys, no server-side storage involved. Try it on

Don’t fight your docs. Just talk to them. We built a document chat app that just works. Any file type, any industry. Legal docs, engineering diagrams, clinical notes and everything in between. Chunkr Chat is now live! See what the best chat feels like. Watch it break down a

this is an awesome use case. good parsing + structured outputs is a such powerful combination. if you don't want to send your data to the cloud, you can also self-host chunkr. we make it easy to spin up a @vllm_project or @ollama container and run an open source LLM.

is it ok to say I love u to your intern chat?.

we love u @OpenRouterAI and @dhrxvb26

It is also open source! We used @supabase as our vector DB, @OpenAI for embeddings and @OpenRouterAI for LLMs. Fork and have fun!.

Building all this with Chunkr is easy. Full step-by-step is in the blog!.

Make your RAG app multimodal. If you build with Chunkr, you can give figures as a cropped image as context to your LLM. This lets your chatbot give users data they can see, not just read about.

Chunkr’s bounding boxes make every source one scroll away. Click a citation and the viewer jumps to the exact paragraph, table cell, or figure.

Level up your RAG app. Most PDF-centered chat experiences fall short on clarity and trust. Our latest blog shows how you can use Chunkr outputs to build standout PDF experiences, with pinpoint inline citations and rendered visuals right inside every LLM response. 🔗 below

RT @piammichel: I didn’t expect to ever say this but this was a fascinating read about PDFs. and how design choices Adobe made in the 90s m….

Read the full post here:

So here we are, building incredibly complex Vision AI to reverse-engineer a 30 year old file format. At Chunkr, we're not just patching the problem. We're building the fundamental infrastructure to turn these static files into a dynamic type for an AI-native world.

Did Adobe try to fix it? Yes. "Tagged PDFs" were meant to add structure, like HTML. But it was a classic chicken and egg problem. No one made them because no one used them. The standard failed, and a multi-billion dollar Document AI industry was born to clean up the mess.

For decades, we built workarounds. But when LLMs arrived, the problem exploded. You can't build reliable AI products on a jumbled mess of text. Garbage in, garbage out. The old tech debt was suddenly a massive blocker for the industry.

But this perfect rendering had a hidden cost: the "flat canvas." Underneath, the PDF was just a painting of the document. It had no concept of headings, paragraphs, tables, reading order etc. For a computer, it was fundamentally unreadable.

It all started with a noble goal. In 1991, Adobe's "Project Camelot" set out to create "digital paper." A file that would look the exact same on any screen or printer. They succeeded, and the PDF became the standard for digital documents.