I built Thinking Canvas, a tool for researching the things you're curious about, whether that's the different types of cotton T-shirts or digitizing handwriting using a pen with cameras on the front. The idea is to have one place for chatting, browsing, and note-taking.
For 1-2 hour exploration sessions, I think this tool is pretty useful! On a Figma-like canvas, you can see chats side by side and fork them. You can choose to browse Google or chat with Claude. You can read webpages and PDFs. If you connect a webpage or PDF to a chat, the chat will have it as context.
The problem is that when you return to a project a day later, it will feel like clutter. There are just too many words and too little signal. It's a Jackson Pollock in the Bauhaus suite.
One problem is that note-taking doesn't really belong here. If I am learning something new and hard, my notes need to be crystal clear and organized, not strewn about with scrappy chats. With AI, we are swimming in lakes of words already, and a canvas makes it feel like a sea where I can't tell friend from foe.
I rooted the notes in the filesystem so I could just open up the folder in Obsidian and read and edit my notes from there. Then, it would be ephemeral chatting and browsing on the canvas and durable note-taking in my notes app. I haven't had the chance to try this though. Theoretically, it's a neat example of symbiosis across apps based on the file-over-app philosophy.
Maybe the bigger problem is that AI has overloaded me with information, and I'm chasing answers to get things done rather than exploring like a student or researcher. So for now, I use Thinking Canvas only rarely for things like fleshing out project ideas.
I have an aspiration to create something really useful that also looks like a toy. This project looks like a toy. But it is only moderately useful.
An interesting strategy for tool makers to just create a frontend on top of an existing agent. Then, users bring their eg Claude subscriptions and perhaps just pay a one-time cost for the frontend. Of course, the labs can shut this down at any time (and I believe they did for OpenClaw).
The design here was influenced by Factorio, which I played for like two days in January. Also by an interactive math playground called Polypad. And, quite strongly, by the Apple Notes icon. And by the greatest of all canvases, Figma. And by the multicolored Post-its on my desk.
The project is open source on GitHub.