timm37

I'm getting hit constantly by bots trying to hack my site, so I'm trying to hack my own sites. Pretty much I have a few obliterated LLMs (DeepSeek-V4.1-Flash-Obliterated, OrcaCyber-Zero, etc) which run on their own sandbox to try to hack me.

If anyone wants to try it, you can test it on mockhack.com

We do verify the request is legit, but yes, same as anyone can hire a pen test, this is all on public endpoints and chances are people are already running them, companies just don't know about it.

Looking at my logs I'm getting waves of daily hack attempts, most are basic scripts, but LLMs are getting more and more common.

I've built a tool which runs a bunch of obliterated LLMs (mainly deepseek, kimi and a few others specialized in cyber attacks), to try to hack my site.

If anyone wants to try it, we can run it and send you a report in about 24h

I'm actually adding "workflows" which allow the platform to ingest files via api, a cusom email address and hosted forms.

You could set a forward rule on your email and auto-forward all those to the [workflow_id]@endtype.email.

If you are interested I can setup an email for you, send me an email at support [at] endtype.com

Most likely you will be able to do this right away.

If not we could easily fine tune a custom model for this task, particularly if you already have a bunch of input/outputs.

Can you send me an email at support [at] endtype.com?

Hey everyone. As AI gets better and better and multimodal I believe one of the most common use cases will extracting structured data from unstructured files. So things like shipping labels, bank statements, invoices, patents, etc.

I plan to release workflows soon which will simply take any file via email or form and save the structured content on a spreadsheet/csv or a new PDF.

Let me know if you would be interested in trying the workflows and if you have a use case to extract/organize different files.

like https://sli.dev ?

Hey HN. I love building things, but it's hard. Currently my options are getting a full time job (and giving up entrepreneurship) or selling a few hours of my time. I'm hoping this works and I can collaborate and help a few companies while being able to work on my projects. Let me know if you have specific questions.

local running would be impossible as you would need your computer running 24/7 (since the widget is available on your website which is always online).

But building a simple RAG/Vector DB on top of llama3 is not very complicated.

Hey, OpenAI recently released "assistants" - These have RAG built into the API, which means that you can provide up to 10.000 files to the assistant without the overhead of dealing with vector databases, splitting files into chunks and more.

They work surprisingly well, so I've built a free simple tool to embed them on websites. It has no dependencies and simply uses fetch to communicate to my backend which proxies openai.

In the future I plan to add more widget embedding options (currently only chat is available).

Let me know if you have any specific questions about either the tool or openai assistants in general.