Love this one! My immediate thought for a build: just a camera with on-device VLMs and LLMs. You could point it at a normal whiteboard (or a wall of sticky notes), and the model could interpret the handwriting, track the tokens, and sync the state digitally in real-time without needing any special "smart" hardware.
It feels like a much more flexible approach than an expensive proprietary whiteboard. Would a camera-only setup cover your use case?
One thing I want: something like a function call graph: you load some source code files, open a function in window A (windows A now shows Only this function with all functions calls and structs highlighted). Click on any function call then the callee shows up in a new window with proper highlighting. Click on a struct object it immediately shows the struct def in a panel stick to the bottom of the screen.
Essentially, it’s like VSCode Peek definition but with a different visual style, and similar to the same functionality of source insight but free and in Linux.
The purpose is to read xv6 source code. I have found some parts of the kernel, e.g. the file system to be convoluted, and I need to follow quite a few jumps to form the system mindset. Having such a small tool is very helpful.
Actually it doesn’t look too hard to implement. Maybe I’ll write it myself in QT.
Although it would be manual, that’s very doable in Emacs and Vim. It’s the very reason I use these editors as you can have a screen fulls of opened files instead of the tabbed model of common editors. One example having all the files you need to edit when you’re implementing a feature in a multilayered architecture. Easier that switching tabs again and again.
I experimented with this idea a couple years ago, using the Language Server Protocol to make it some what universal. The output gets messy pretty quickly, maybe these days an LLM could be used to only show the interesting calls.
Planners. Something to help me make sense of what my priorities are during the day. Something to help me determine what needs my attention.
I have a lot of systems, but I could really use a Jarvis at work right now and it seems like the whole damn promise of AI is to deliver that. I’m waiting.
There's something about having so many planners online/digital that sucks.
I want to plan, but I don't want to be on the computer yet. But if I plan non-digitally, it's slower, I can't copy/paste, and then I don't really have good access to it unless I have it with me.
But if I plan too much digitally, it's stored in digital memory, not my memory.
I'm really struggling to find something to mitigate this. I wish I had a tactile miro board that also created the miro board online.
An LLM tool that can sit on a CI pipeline to propose what tests should be blocking.
Instead of brute-force method of selecting the appropriate test suites by path or similar, have LLM analyze changes and propose the set of test suites that is relevant to the change.
If there are new complex tests added to the change, estimates how many times to run them to ensure they are not flaky to begin with (hundreds? or thousands?).
A physical board that translates digitally.
Imagine a whiteboard that has sticky notes, writing, little tokens and trinkets and the board also becomes a digital version that you can iterate on.
I really like to plan with my hands and in MY memory, but still love the utikity of planning digitally of course.
Love this one! My immediate thought for a build: just a camera with on-device VLMs and LLMs. You could point it at a normal whiteboard (or a wall of sticky notes), and the model could interpret the handwriting, track the tokens, and sync the state digitally in real-time without needing any special "smart" hardware.
It feels like a much more flexible approach than an expensive proprietary whiteboard. Would a camera-only setup cover your use case?
One thing I want: something like a function call graph: you load some source code files, open a function in window A (windows A now shows Only this function with all functions calls and structs highlighted). Click on any function call then the callee shows up in a new window with proper highlighting. Click on a struct object it immediately shows the struct def in a panel stick to the bottom of the screen.
Essentially, it’s like VSCode Peek definition but with a different visual style, and similar to the same functionality of source insight but free and in Linux.
The purpose is to read xv6 source code. I have found some parts of the kernel, e.g. the file system to be convoluted, and I need to follow quite a few jumps to form the system mindset. Having such a small tool is very helpful.
Actually it doesn’t look too hard to implement. Maybe I’ll write it myself in QT.
Although it would be manual, that’s very doable in Emacs and Vim. It’s the very reason I use these editors as you can have a screen fulls of opened files instead of the tabbed model of common editors. One example having all the files you need to edit when you’re implementing a feature in a multilayered architecture. Easier that switching tabs again and again.
I experimented with this idea a couple years ago, using the Language Server Protocol to make it some what universal. The output gets messy pretty quickly, maybe these days an LLM could be used to only show the interesting calls.
Planners. Something to help me make sense of what my priorities are during the day. Something to help me determine what needs my attention.
I have a lot of systems, but I could really use a Jarvis at work right now and it seems like the whole damn promise of AI is to deliver that. I’m waiting.
There's something about having so many planners online/digital that sucks.
I want to plan, but I don't want to be on the computer yet. But if I plan non-digitally, it's slower, I can't copy/paste, and then I don't really have good access to it unless I have it with me.
But if I plan too much digitally, it's stored in digital memory, not my memory.
I'm really struggling to find something to mitigate this. I wish I had a tactile miro board that also created the miro board online.
Temporary ssh containers, I just want to ssh (some id)@(container host), spin up a lxc container or firecracker microvm, it attaches to it.
When I exit the ssh session, it terminates (or "pauses").
I've wanted to make it for a while but never got around to it.
You can get such disposable shell in k8s with a command like:
kubectl run -i --tty --rm debug --image=busybox --restart=Never -- sh
Doesn't use ssh though.
Unless I'm missing it, the linuxserver have a container you can use. There's also containerssh (which I use all the time)
https://hub.docker.com/r/linuxserver/openssh-server
https://hub.docker.com/r/containerssh/containerssh
Oh interesting! ContainerSSH looks like what I want, thanks.
Unbloated easy to use postman.
Is a terminal app for you? https://posting.sh/ (You can also use it with the mouse)
have you tried this one: https://github.com/usebruno/bruno
Still bloated i myself am working on https://voiden.md
An LLM tool that can sit on a CI pipeline to propose what tests should be blocking.
Instead of brute-force method of selecting the appropriate test suites by path or similar, have LLM analyze changes and propose the set of test suites that is relevant to the change.
If there are new complex tests added to the change, estimates how many times to run them to ensure they are not flaky to begin with (hundreds? or thousands?).