Hi HN, I’ve been experimenting with this format for a while, and I think I’m ready to share what I’ve built.
I use it quite a lot myself and have learned a bunch of cool stuff with it. I usually start with a general question like “how does X work” and go deeper and deeper. AI is actually pretty good at explaining complicated topics if you give it the right medium to work with.
So I’ve been exploring physics, biology, geometry, etc. I’ve never been so hungry for knowledge and so curious about things in my life (I think). I know ChatGPT already gives that feeling in a way, but this is somehow different. I would open ChatGPT voice and work with it on a board for hours and hours, forgetting about my work.
You can ask your agent to research and explain pretty much anything, or ask it to give you exercises to work on. It has instructions not to give away answers when you’re working on a problem, and to explain things step by step on the board like a real teacher would.
The app
It’s currently a macOS app. I’ll add Windows and Linux soon if people ask. It’s free and open source, so everyone is welcome to contribute.
Your agent (whatever AI subscription you have) connects to it over MCP. It can view the board and draw on it with sandboxed JavaScript and simulated ink. It can also place three.js views, HTML and SVG drawings. It gets updates on what changed on the board, so it can see your work.
I use it with a drawing tablet, but a mouse and keyboard work just as well.
I set up a server with Gemini for the voice, so it’s easier to start. It should give about 55 minutes per user per day, but if a lot of you show up, my credits will run out in a minute. So if there’s no voice for some reason, open Settings and install the local model. It’s just as good, but it’s a separate install because of licensing and size.
So yeah, give it a try! Can’t wait to see your board recordings (I think I need to make that feature).
Btw, Opus 5.5 on medium/high is the best explainer so far, and try it with ChatGPT voice.
I have some thoughts on how to make this work 10x better, more on the harness and agent interaction side.
Haven't tested this with ChatGPT Ultrafast mode. Can't wait to get the $500 plan. Or whenever that becomes more available. Because speed matters here even more.
Thanks! I think it has to be more accessible than the current setup for students. I built this for people who already use AI beyond just simple chat. Wanted to test whether this is the right direction in the first place.
Already working on a web version where you just open the page and it will work.
Hi HN, I’ve been experimenting with this format for a while, and I think I’m ready to share what I’ve built.
I use it quite a lot myself and have learned a bunch of cool stuff with it. I usually start with a general question like “how does X work” and go deeper and deeper. AI is actually pretty good at explaining complicated topics if you give it the right medium to work with.
So I’ve been exploring physics, biology, geometry, etc. I’ve never been so hungry for knowledge and so curious about things in my life (I think). I know ChatGPT already gives that feeling in a way, but this is somehow different. I would open ChatGPT voice and work with it on a board for hours and hours, forgetting about my work.
You can ask your agent to research and explain pretty much anything, or ask it to give you exercises to work on. It has instructions not to give away answers when you’re working on a problem, and to explain things step by step on the board like a real teacher would.
The app
It’s currently a macOS app. I’ll add Windows and Linux soon if people ask. It’s free and open source, so everyone is welcome to contribute.
Your agent (whatever AI subscription you have) connects to it over MCP. It can view the board and draw on it with sandboxed JavaScript and simulated ink. It can also place three.js views, HTML and SVG drawings. It gets updates on what changed on the board, so it can see your work.
I use it with a drawing tablet, but a mouse and keyboard work just as well.
I set up a server with Gemini for the voice, so it’s easier to start. It should give about 55 minutes per user per day, but if a lot of you show up, my credits will run out in a minute. So if there’s no voice for some reason, open Settings and install the local model. It’s just as good, but it’s a separate install because of licensing and size.
So yeah, give it a try! Can’t wait to see your board recordings (I think I need to make that feature).
Btw, Opus 5.5 on medium/high is the best explainer so far, and try it with ChatGPT voice.
Repo: https://github.com/Nordup/askaboard
I have some thoughts on how to make this work 10x better, more on the harness and agent interaction side.
Haven't tested this with ChatGPT Ultrafast mode. Can't wait to get the $500 plan. Or whenever that becomes more available. Because speed matters here even more.
Very interesting idea, It can be very beneficial for students using AI for studies.
Thanks! I think it has to be more accessible than the current setup for students. I built this for people who already use AI beyond just simple chat. Wanted to test whether this is the right direction in the first place.
Already working on a web version where you just open the page and it will work.