This is quite interesting, let me ponder over this a bit. I think I can expose an MCP and also make the admin to be hosted directly in cloudflare workers/pages and make it a remote MCP.
Will need to handle auth though since the dashboard will be public.
Yes, the readme and no screenshot was something that I meant to cleanup and add and are already live now.
Agent reports vs small static sites -> It was supposed to mean markdown files as plans gets shared between teams which is what CC mentioned as agent reports.
Yes, code is definitely cheap. Built it as an utility for myself and dogfooding it with my team. Had put it out with the thought that others might find it useful.
Point in time, a lot of it is available vanilla with cloudflare cli but natively enabling passwords, private team dashboard are something that am considering. Passwords are already live as we speak.
Will improve this as i get more feedback and time to spend on this.
publishing workflow yes, i wanted to have a way to manage published urls and update existing urls with a small dashboard as well since we end up doing it repeatedly.
Just noticing this, i have covered all the rules in the guide. It's not as complicated as Mahjong. Much simpler and as mentioned earlier it is regional to south India
Gitleaks is too good to be compared to this, the only areas where keyleak is comparable is it does runtime detection where sometimes your build process injects secrets via env vars into the bundle or any of your responses expose a config file or secrets.
Yep, Github does a wonderful job flagging secrets most times but irrespective of that this is happening in some sites. This was built out of personal curiosity and I had put it out for public because I myself was not aware if this will be actually useful and if so in what form. People do YOLO to prod, we have more kids building AI wrappers than I can count, and somewhere in that chaos secrets slip through the cracks.
Ideally I would have loved this to be a chrome plugin or part of the CI/CD pipeline or put it out as an adversary agent for all of these new vibe coded apps but don't think I'm that vested into the idea yet. Thanks for being gentle :)
Well, when i meant "personally" not in the app I manage. I have a quirk of checking sites to understand what they are using and how they are using and have stumbled upon sites with exposed Gemini, Google Maps, OpenAI keys etc.
With all the vibe coded apps that are getting launched or were launched early, there are enough holes to plug. This is just an attempt to help individuals or orgs to ensure they are not exposed. Just pushed it out what I had in mind based on my experience.
And I agree with you that an adversary approach won't work if we can't fix the underlying problem but the world has changed with enough vibe coded apps that are getting shipped everyday and very little of them care or know about security.
Wanted to pen down some of my thoughts on ambition and how we seek validation and can it be abstracted to a certain degree so that we can break free from it
Just stumbled upon this blog which details out testing and validating large GPU clusters before running training workloads. Any other similar blogs which adds more nuance in terms of debugging the issues once we identify them as well?
I just built a simple chrome extension which uses Groq with a whisper model to do Speech to Text so that we can give speech inputs to Claude. I really liked being able to do speech to text with ChatGPT and missed that quite a bit with Claude, hence built this.
Why Groq?
They have a free tier and super fast, so anyone can create an account and add their API key to the extension.
Privacy?
The only API calls that are there are to Groq servers, so as long as you have read their T&C you should be fine
This was done and tested in a few hours and quite buggy still but works. Feel free to drop some feedback. Will push more updates, if anyone else actually finds this useful
Just started working on this small project where one can just convert their whatsapp conversations into an instruction dataset which can be finetuned with any OSS models for creating a clone of themselves.
Have added sensitive info redaction and also general whatsapp messages redaction when parsing and compiling the dataset.
Very early but appreciate any suggestions or discussions around this.
I was just dumping all of my thoughts with respect to LLMs on how to setup and manage a RAG system. Thought it will be useful for people to think through as well, hence sharing here.
you can check it out here: https://github.com/Amal-David/pagecast