Show HN: Site2pdf (opens in new tab)

(github.com)

5 pointslaiso1y ago8 comments

Hello everyone, I created a tool called "site2pdf" over the weekend. This tool converts the main page and sub-pages of a website that match a specified URL pattern into a PDF file. It is particularly suitable for AI-based RAG (Retrieval-Augmented Generation) and QA (Question Answering) tasks.

GitHub: https://github.com/laiso/site2pdf/

# Features

- Generate PDFs of main and sub-pages

- Based on Node.js and Puppeteer

- Easy-to-use CLI tool

I want to make this software available online for my friends, but I'm struggling with the best architecture to use. I want to meet the following requirements:

- Cost-effective

- Use Cloudflare Workers' Browser Rendering API(Managed Puppeteer)

- Save to Workers Queue -> R2 bucket

I have already created a prototype, but it encounters ExceededCpu errors when running the consumer for a long time. It seems I need to implement a distributed architecture including merging, which seems challenging. I would appreciate any advice you can give. Thank you!

I look forward to your feedback!

8 comments

miles1y ago

Thank you for crafting and sharing this.

You mention that the generated PDFs are "particularly suitable for AI-based RAG and QA tasks" - can you please share your preferred method/tools for that?

laisoOP1y ago

I particularly like using Google NotebookLM, as it allows me to consolidate books and documents in one place for knowledge search.

https://github.com/user-attachments/assets/3c68298e-265f-410...

Additionally, I have created GPTs that can have conversations about Tauri using PDFs as the source.

https://chatgpt.com/g/g-Pa0nP2mJX-tauri-v2-helper

miles1y ago

Thanks for taking the time to respond. I was thinking of something local, especially in light of:

Google's Gemini AI caught scanning Google Drive PDF files without permission https://news.ycombinator.com/item?id=40965892 .

Looks like GPT4All[1] and AnythingLLM[2] are worth exploring. There's also the closed-source macOS app RecurseChat[3,4] which appeared on HN a few months ago[5].

[1] https://github.com/nomic-ai/gpt4all

[2] https://github.com/Mintplex-Labs/anything-llm

[3] https://recurse.chat

[4] https://recurse.chat/blog/posts/local-docs

[5] https://news.ycombinator.com/item?id=39532367

1 more reply

bosch_mind1y ago

can you elaborate on the execution failures?

If you’re referring to single pages taking too long and wanting to chunk the processing of a single page, consider cloudflare durable objects as a coordination point for a url.

You can parallelize chunks and track completion in DO. When the last chunk is complete, you can merge. Unlike KV, DO is strongly consistent.

laisoOP1y ago

I have conducted further investigations following this post.

https://github.com/laiso/site2pdf/issues/6

The merge process using only KV has been successful. The real issue lies in the concurrency limits of Browser Rendering API.

unstatusthequo1y ago

Is there a way to simply feed it a list of known URLs directly without it trying to figure out the sub-pages? Complex and large site discovery can be a pain.

laisoOP1y ago

Currently, site2pdf does not support feeding it a list of known URLs directly. However, I believe detecting sub-pages from a sitemap.xml file could work well. Thank you for the question!

j / k navigate · click thread line to collapse

Show HN: Site2pdf (opens in new tab)

(github.com)

5 pointslaiso1y ago8 comments

GitHub: https://github.com/laiso/site2pdf/

# Features

- Generate PDFs of main and sub-pages

- Based on Node.js and Puppeteer

- Easy-to-use CLI tool

I want to make this software available online for my friends, but I'm struggling with the best architecture to use. I want to meet the following requirements:

- Cost-effective

- Use Cloudflare Workers' Browser Rendering API(Managed Puppeteer)

- Save to Workers Queue -> R2 bucket

I look forward to your feedback!

8 comments

miles1y ago

Thank you for crafting and sharing this.

You mention that the generated PDFs are "particularly suitable for AI-based RAG and QA tasks" - can you please share your preferred method/tools for that?

laisoOP1y ago

I particularly like using Google NotebookLM, as it allows me to consolidate books and documents in one place for knowledge search.

https://github.com/user-attachments/assets/3c68298e-265f-410...

Additionally, I have created GPTs that can have conversations about Tauri using PDFs as the source.

https://chatgpt.com/g/g-Pa0nP2mJX-tauri-v2-helper

miles1y ago

Thanks for taking the time to respond. I was thinking of something local, especially in light of:

Google's Gemini AI caught scanning Google Drive PDF files without permission https://news.ycombinator.com/item?id=40965892 .

Looks like GPT4All[1] and AnythingLLM[2] are worth exploring. There's also the closed-source macOS app RecurseChat[3,4] which appeared on HN a few months ago[5].

[1] https://github.com/nomic-ai/gpt4all

[2] https://github.com/Mintplex-Labs/anything-llm

[3] https://recurse.chat

[4] https://recurse.chat/blog/posts/local-docs

[5] https://news.ycombinator.com/item?id=39532367

1 more reply

bosch_mind1y ago

can you elaborate on the execution failures?

If you’re referring to single pages taking too long and wanting to chunk the processing of a single page, consider cloudflare durable objects as a coordination point for a url.

You can parallelize chunks and track completion in DO. When the last chunk is complete, you can merge. Unlike KV, DO is strongly consistent.

laisoOP1y ago

I have conducted further investigations following this post.

https://github.com/laiso/site2pdf/issues/6

The merge process using only KV has been successful. The real issue lies in the concurrency limits of Browser Rendering API.

unstatusthequo1y ago

Is there a way to simply feed it a list of known URLs directly without it trying to figure out the sub-pages? Complex and large site discovery can be a pain.

laisoOP1y ago

Currently, site2pdf does not support feeding it a list of known URLs directly. However, I believe detecting sub-pages from a sitemap.xml file could work well. Thank you for the question!

j / k navigate · click thread line to collapse