Launching today
SayCraft
Build a web app by talking through a meeting
43 followers
Build a web app by talking through a meeting
43 followers
SayCraft turns a real-time team meeting into a working web app. You talk — alone or with your team — the AI builds live, and a shareable preview URL updates as you speak. Walk away with a working product, the source code, and one-click deploy. No prompts.
Turning a live meeting into a build session with a preview URL that updates as you talk is a genuinely different loop from prompt-by-prompt, and the no-prompts claim is the part I'd want to stress-test. When two people talk over each other, how does it decide whose intent to build, and can I correct or roll back a wrong inference mid-meeting without stopping the session? Also, does the source you walk away with come out as a normal repo I own and can keep iterating on in my own editor, or is it tied to your platform to stay runnable?
@noctis06
Great questions — this is exactly the part worth stress-testing, so let me be specific.
Two people talking over each other / conflicting intent: it doesn't quietly pick a "winner" behind your back. When there's a real disagreement, you can just have it build both directions and actually try each one live, then decide by feel instead of arguing about it in the abstract. And the session replay keeps every intermediate state, so you can scroll back through every form it passed through along the way — nothing gets lost.
Correcting a wrong inference mid-meeting: yep, no need to stop. You just say it — "no, go back to the simpler header," "make it the other version" — and the UI updates live as you talk. Course-correcting by speaking is the loop; there's no restart, no re-prompting.
The code you walk away with: completely platform-independent. It's a normal, complete, genuinely runnable codebase you own — clone it, open it in your own editor, keep iterating. Nothing's tied to us to stay runnable.
Honestly the best way to test the "no-prompts" claim is to run a real session and try to break it — would love to hear how it holds up for you.
For anyone who'd rather see it than take my word for it — here are a couple of apps built entirely by talking through a meeting, each with the live app and the full session replay so you can watch the conversation turn into the product:
🌦️ Skycast — a weather app, dark cyberpunk theme
• Live app → https://emxtbkvilzv.flashdesign.tech
• Watch it get built (full meeting replay) → https://saycraft.ai/share/Ppj0zpSaVxf
🪐 Helios — an interactive 3D solar-system explorer
• Live app → https://elzgw4qyo3e.flashdesign.tech
• Watch it get built → https://saycraft.ai/share/By4GWlfmDx6
The replay is the fun part: scrub through the actual session and watch the preview update sentence by sentence as the team talks. No prompts typed anywhere.
Ame, this is a fascinating approach to the build process. Sitting in endless client discovery sessions and architectural meetings, the hardest part is always translating spoken requirements into functional code without losing context. Building it during the conversation is a massive paradigm shift.
Since the codebase is fully exportable, I am really curious about the underlying stack it generates. Is it primarily outputting a React/Next.js frontend, or does it also attempt to scaffold out the backend APIs and database schemas if the team starts discussing data structures? Huge congrats on the launch!
@varunvivek Thanks so much — really appreciate the support. And yeah, that "translating spoken requirements into code without losing context" pain is exactly what got me building this.
Straight answer on the stack: right now it outputs a **React frontend web app, and that's it.** Today the focus is turning a live conversation into something you can *see and click* in real time, so it runs on mock data rather than real persistence. If the team starts talking data structures, it'll model them in the UI — lists, forms, the shapes of things — but it won't scaffold real backend APIs or a database yet.
That's the deliberate v1 scope: the fast "talk → watch it appear" loop for **real-time demos and prototypes**. Real backend / API / schema generation is squarely on the roadmap as it grows from a demo tool into something you'd ship end-to-end. And since the output is a clean, exportable React repo, you can always wire your own backend onto it in your editor in the meantime.
Thanks again for the thoughtful question — exactly the kind of feedback that helps me prioritize what's next ！
Voquill
The live collaboration angle is interesting and the no prompts approach is a fresh take. How much control is there over architecture and code quality as projects get more complex? Congrats and good luck!