We're live: https://t.co/hKTIsbKbZo 🍜
What the world actually cooks. Any dish, 12 languages, and it even reads menus you can't.
Huge inspiration: @heystefan_@theadistar@marclou 🙏
An agent now picks what I post about. It only reads my last 30 days of commits. Old projects stay quiet.
This post was planned that way. I still approve every one.
Where would you draw the line: topics, drafting, or posting?
You have eggs, rice and half an onion. What can you cook?
Stackplate suggests dishes from what you have. Ones you can fully make come first. Salt and pepper don't count against you.
Which 3 ingredients are in your fridge? I'll reply with what it suggests.
What does the world actually cook at home?
A globe where every country wears its flag. Turn it, tap a country, and get its home dishes: the ones families make, not the restaurant menu. Tap one and the recipe opens.
Looking for one place? Type its name.
My AI team runs my X. New rooms below: YouTube and Instagram offices.
YouTube: idea, script, cover. 3 Shorts per video.
Instagram strategy agent: real data, costs less.
Built in public, early. 10 commits in this update.
One project, 3 channels: which step stays human?
You don't need the English spelling.
Type a dish the way you know it: ผัดไทย, 味噌ラーメン, Борщ, كبسة. Not in the kitchen yet? The recipe is written while you watch.
Then read it, and other cooks' comments, in your language.
Most AI apps call the big model first. I call it last.
Stackplate checks a recipe library first, then a cache, then a small model. The LLM only runs when nothing else has the answer.
Cheap path first, smart path as fallback. How do you keep AI costs down?
AI agents run this X account. I watch them in a pixel-art office.
Working system in one day. 15 commits.
Weekly report with real X metrics.
No results yet. Day one, expect mistakes. At least no raises.
Which job would you never hand over? Ask me about the setup.
@nadeematx Redis, yes. Upstash on Vercel. But most requests never hit it: a hand-checked library answers first, then the Redis cache, and the LLM only runs when both miss. Which reads are slowest for you right now?
Traveling and can't read the menu?
Snap it. Every dish comes back translated, with how to say it, the price and what's in it. Then save the whole menu as one image and show it to the waiter.