ThreadChat — Restructuring AI Conversations
What Inspired This Project
Every day, I use ChatGPT for learning — algorithm explanations, code walkthroughs, concept clarifications. But there's a friction point that's always bothered me:
I get a detailed answer. Halfway through reading, something confuses me. I ask a follow-up question. Then another. And another. Soon my conversation is a long linear thread, and going back to the original answer requires scrolling all the way up.
The problem: AI chatbots are optimized for linear chat. But thinking isn't linear.
When I encounter a doubt, I want to explore it right there — without losing my place in the original conversation. I want to see my entire exploration later — not as a scroll-heavy wall of text, but as a map.
That's when I realized: every answer should be a branching point. Not just the prompts — the answers themselves.
How I Built It
I turned to MeDo and described my vision entirely in natural language. No hand-coding. No wrestling with databases or backend plumbing.
The Core System
For the branching tree system — the structural backbone of BranchChat — I gave MeDo:
- A brief description of the three-level logic (main conversation → branch chats → sub-branches)
- A rough reference image of what I had in mind
MeDo understood it immediately and generated the complete system on its own. The data model, the relationships between parent and child conversations, the context passing — all from a single description.
The Tree View
The most impressive part: the visual tree map. I described it as a mind-map where each answer is a node and branches are its children.
MeDo built a fully working interactive component from that single description. Users can now see their entire conversation as a visual hierarchy — the main trunk, every branch exploration, all at a glance. No scrolling needed.
Authentication & Guest Access
For the login system, I used MeDo's Login skill plugin. I typed one line:
"Implement an authentication system with a guest login option"
MeDo handled the complete auth flow, user persistence, and guest session management. One line of instruction. Full production-ready login system.
The Side Panel Chat
When a user clicks an answer, a side panel slides in from the right. The panel shows the parent answer (read-only, for context) and a fresh chat input below. The AI in that branch is aware of the parent answer as its context — it knows exactly what it's clarifying.
This multi-turn conversation capability in MeDo made this possible. Each branch maintains its own conversation history, all tied back to the parent.
What I Learned
1. MeDo's Natural Language Comprehension is Remarkable
I expected to need detailed wireframes, API documentation, database schemas. Instead, I described things in plain English — sometimes vaguely — and MeDo translated intent into working code. This freed me to focus on design and UX rather than implementation details.
2. No-Code Doesn't Mean Dumbed-Down
I was skeptical that a no-code tool could handle a non-trivial data model (nested conversations, parent-child relationships, multi-turn context awareness). But MeDo built it all without me writing a single line of code. The platform is production-grade, not a prototype builder.
3. Iteration is Faster When You're Not Coding
If I'd hand-coded this, a single refinement (e.g., "add guest login") would mean touching multiple files, running migrations, testing. In MeDo, I just iterated through a conversation. Refinements took minutes, not hours.
4. Plugins + Natural Language = Powerful
Using the Login skill plugin wasn't just convenient — it validated that MeDo's plugin ecosystem works seamlessly with natural language prompting. I didn't read API docs or install packages. I described what I wanted, and MeDo connected the dots.
Challenges I Faced
1. Scoping for a 48-Hour Hackathon
My initial concept included features like:
- Persistent storage of branch conversations
- Export to Markdown/PDF
- Collaborative branching (multiple users exploring the same tree)
- Advanced analytics on which branches users explore most
I had to cut ruthlessly. The MVP is a fully functional branching chat with a tree view. Everything else is v2.
Lesson: A tight, working MVP beats a sprawling half-built feature set. Judges would rather see depth on core features than breadth with gaps.
2. Context Passing in Branches
Early iterations, branch chats lost context of the parent answer. The AI would give generic clarifications instead of answers rooted in the original response.
Fix: I refined the prompt in MeDo to pass the parent answer as explicit system context in every branch chat. This took a few iterations, but MeDo's multi-turn conversation made it easy to test and refine.
3. Tree View Rendering
Getting the tree visualization to be both clear and interactive was tricky. Too compact, and overlapping branches are hard to read. Too spread out, and it doesn't fit the screen.
Fix: MeDo's component generation handled most of the layout, but I had to give it specific guidance on spacing and hierarchy depth. A few conversational refinements got it right.
4. Mobile Responsiveness
The side panel chat + tree view + main chat felt cramped on mobile. Horizontal scrolling was inevitable, which defeats the "no scrolling" value prop.
Partial fix: The tree view collapses on mobile. The side panel chat takes full width. It's a compromise — not perfect, but functional. A full mobile redesign is on the roadmap.
The Differentiation Angle
ChatGPT added branching in 2025. But theirs branches from prompts — you need to know what to ask before you diverge.
BranchChat branches from answers — because doubts happen while reading, not before. You're exploring an answer, something confuses you, you click and dive deeper. No scrolling. No losing your place.
That distinction — answer-level branching + a tree map of your entire exploration — is what makes BranchChat unique.
What's Next
- Persistence: Save conversation trees to the user's account so they can revisit explorations later
- Export: Generate a clean markdown or PDF of any branch exploration
- Mobile-first redesign: Full responsive tree view and side panel
- Keyboard shortcuts: Power users should be able to branch and navigate without touching the mouse
- API: Let other apps embed BranchChat as a widget for learning experiences
Final Thoughts
This hackathon taught me that the best ideas often come from personal friction. I wasn't trying to solve a theoretical problem — I was solving my own daily frustration with linear chat interfaces.
MeDo made it possible to go from idea to working MVP in 48 hours. Not a prototype. Not a mockup. A real, deployed, usable product.
If you've ever felt lost in a ChatGPT conversation, or wished you could explore doubts without losing context — BranchChat is for you.
TL;DR: I was frustrated scrolling endlessly in ChatGPT. I built a branching chatbot where every answer spawns explorations, and your entire conversation becomes a visual tree. Built entirely with MeDo, no code written by hand. One tagline sums it up:
Branch, don't scroll. learned
What's next for ThreadChat
Built With
- medo
