I think most apps should look like this, vastly simplified, do one thing...

There's a crazy paradox between how powerful AI/LLMs can be, and that one can just "steal the weights." If AI gets really, really powerful, the amount of resources that malicious actors put into obtaining those weights will match. Newsletter in bio!

Nice to see Canadian companies!!!

Need to release a new version of raillock, so need a commit message, hey, why not use an AI generated commit message...now I have two problems! Trying out Open Commit

I don't know enough about packages in the nextjs vercel world, so am working on a little script to show me more info. As I was working on that that, Node.js released a new version of the LTS release, ha! Out of date already, within 10 minutes of updating

More work on raillock, mcp tool checksumming It's a CLI and it's a library that agents can use to validate mcp tools In this image, I'm comparing a previously created config file with a running mcp server (that happens to have malicious tools)

It's a good time to be a Linux user, Cursor, Windsurf, Zed...they all have Linux clients

well, finally really using this 3090 GPU gwen 32b fully pinned, fan sounds like helicopter one reason it's pinned is because if I get bad json from it, I just send it right back to ask it to fix it, which seems to be happening a lot

This is pretty smart We need ways to find prompt injection, and regexes or yara rules aren't going to be able to do everything, so we'll need things like this nova tool

A new beta release of RailLock is now available, featuring an embedded web GUI. There's still more to do, but it's starting to take shape! Raillock reviews and 'locks' MCP tool descriptions, so if they change, you'll know. It is also a library that can be imported into Python