Framer AI AgentsDesign and publish professional sites with AI
Promoted
Launching today
ThinkDeepWide
Give every question its own branch to help you think.
4 followers
Give every question its own branch to help you think.
4 followers
ThinkDeepWide is for deep thinkers who want to learn and grow with AI, without giving up their own thinking. Every answer brings new questions, and long chats make you lose track. The app turns your chats into a tree, so you can dig deeper and return to earlier questions. Its built-in agents can read files, edit and run code, and work in parallel. The app optimizes prompt caching to help you save money on tokens. You can use your existing Codex or Grok subscription right inside ThinkDeepWide.
Hi Product Hunt! 👋
I’m Roc, one of the makers of ThinkDeepWide. I’m sharing it as part of the GPT-6 Astra Challenge.
Before I introduce ThinkDeepWide, I'd like to share why am i building it.
Motivation
AI has become much more powerful in recent years, which sounds like a wonderful thing.
However, after being immersed in vibe coding for half a year, I realized I was getting lazier and lazier about thinking for myself!
The reason is: AI is smarter than me in coding and many other fields, it is hard to keep up with what it’s doing and understand why, since it knows far more than I do.
And the result is: AI can work things out but I am not getting any better!
So I started looking at how I was using AI.
I noticed that when I ask AI about something I don’t understand, its answer often gives me more questions.
Now, here’s where I get stuck.
If I ask them all at once (Think wide), AI gives a long answer that covers a little of everything, but I still don’t understand each part well enough.
If I focus on one question (Think deep), and keep discussing it with AI through many follow-ups, yes, I will eventually understand it. But by that time, I’ve lost track of the other questions I wanted to ask.(Scrolling all the way back is frustrating)
I bet some of you have experienced this too, and I don’t think the problem is that we don’t want to think deeply.
A long, linear conversation just makes it hard to keep track of our thinking.
That’s why I’m building ThinkDeepWide.
The solution
Give every question its own branch.
ThinkDeepWide turns your conversation into a tree. You can explore one question through multiple follow-ups, then click an earlier node and take another direction. Your other discussions are still there when you come back.
You can also:
Keep asking while AI is still working. Different branches can respond in parallel.
Keep token costs down. We optimize cache reuse so new branches can reuse cached input from ancestor nodes.
Bring your discussions together. Each node knows what’s being discussed in other branches, so you can ask the agent to organize them into notes you can return to later.
And you can get work done in the same app.
ThinkDeepWide is a desktop AI agent: it can read local files, search the web, edit code, and run it.
The way I want to use it is simple: understand the problem with AI, work out a plan together, then let the agent carry it out. I want to learn through that process and not outsource all my thinking.
How Astra helped me build it
ThinkDeepWide builds on BB Desktop and customized versions of the Codex and Grok engines.
Some early features were built with GPT-5.6 Sol. When Astra arrived, two things really stood out to me.
First, its computer-use capabilities were much stronger. It was much better at testing the interface and finding UI bugs. The chat tree is a key part of ThinkDeepWide, so this made a big difference in building and improving the app.
Second, Astra was especially good at planning and directing the work of multiple agents. After I worked out a plan with it, Astra acted as the orchestrator and directed other agents working on the app, frontend, and backend.
Codex also made it easier to keep working when I was away from my desk. I could continue my sessions from my phone while walking or riding the train. That saved me a lot of time.
I still spent a lot of time testing the app, refining the UI, working out how to reuse the cache and so on. With Astra, I could do much more while staying involved in those decisions.
More about the app
The app supports macOS and SSH access to Linux hosts. My teammates are adapting it to the Windows version.
You can use your own ChatGPT or Grok subscription to use the agents in ThinkDeepWide.
You can try the guided interactive and free exploration demos and watch the walkthrough video on our website.
👉 Try the guided demo on our webiste ( best viewed on a desktop or laptop)
Do you run into the same problem? Does exploring one question with AI make you lose track of the others you wanted to ask?
After trying the demo, do you think it would help you? I’d love to hear where you would use it—and what still feels awkward or missing.
Thanks for taking the time to read this long comment!
I know it’s a long read in a world of short videos and quick posts.