What To Eat Review: The 3.3k-Star AI Recipe Generator That Makes Cooking Fun

One of life’s most annoying daily questions: what should I eat today?

Sounds trivial, but if you take it seriously, there’s a lot to consider — what ingredients are in the fridge, what cuisine you’re craving, nutritional balance, and whether you can actually cook it. My old solution was rotating through the same dozen dishes until I got sick of all of them.

Recently found an interesting project on GitHub: liu-ziting/what-to-eat. 3.3k stars, built with Vue 3 + TypeScript, tagline is “one meal to rule them all.” Uses AI to generate recipes. Sounds fun, so I tried it out.

What problem does it solve

One core thing: use AI to solve “what should I eat” and “how do I make it.”

Not just a simple recipe search — where you type “tomato egg” and get a few recipes. More like an AI cooking assistant:

• Tell it what’s in your fridge, it suggests a few recipe options
• Specify a cuisine like Sichuan or Japanese, it generates recipes in that style
• It analyzes nutritional content for each dish
• It recommends wine pairings
• It even generates AI “food photos” — a bit fake-looking but definitely appetite-inducing

The quirkiest feature is “food fortune-telling.” When you can’t decide what to eat, click a button and AI randomly picks a dish for you with instructions and tips. For someone like me with decision paralysis, this is a lifesaver.

Core features

Smart recipe generation Input ingredients or a desired cuisine, and AI generates a complete recipe with:

• Ingredient list and quantities
• Detailed step-by-step instructions
• Cooking time and difficulty rating
• Tips and notes for each step

I tried “potato, beef, onion, Chinese style” and got a beef and potato stew recipe with legit steps — from blanching to caramelizing sugar to reducing the sauce. Unlike some AI-generated recipes that are pure nonsense.

China’s eight major cuisines + international Broad coverage: Sichuan, Shandong, Cantonese, Jiangsu, Fujian, Zhejiang, Hunan, Anhui, plus Japanese, Korean, Western, and Southeast Asian. Each cuisine has corresponding flavor adjustments, and generated recipes reflect their distinctive characteristics.

I tested generating chicken dishes in both Sichuan and Japanese styles. The Sichuan version emphasized spicy and heavy seasoning; the Japanese version was lighter and more ingredient-focused. The distinction was clear.

Nutrition analysis Every generated recipe includes nutritional breakdown: calories, protein, fat, carbs, fiber. Useful for people watching their diet — you can see exactly what you’re getting.

AI food images AI-generated photos of the finished dish. Honestly — the photos look great but may not match what you actually produce. Still, better than nothing for social media or food blogs.

Wine pairing Generated recipes include drink recommendations: red wine, white wine, sake, beer. I tested a few times and the suggestions were reasonable — braised pork with yellow wine, salmon with sake, basically common-sense pairings.

Sauce design This is a neat feature — it auto-generates a matching sauce recipe for the main dish. Steak gets a black pepper sauce; steamed fish gets a soy sauce blend. Small details that genuinely help level up your cooking.

Dynamic configuration Switch AI providers without restarting the service. Supports any OpenAI-standard API: OpenAI, Claude, Gemini, or self-hosted models. Click the settings button, enter your API key and endpoint, and it works immediately.

I tried different models: GPT-4 gave the most detailed recipes, Claude was more creative, and domestic models were slightly simpler in instructions. The flexibility is genuinely useful — switch models based on your API budget.

Real-world usage

Scenario 1: Clear the fridge Random leftover ingredients and no idea how to combine them. Feed the list to what-to-eat, it generates a few recipe options, pick whichever looks good.

Scenario 2: Learn new cuisines Want to learn Sichuan cooking but don’t know where to start. Specify Sichuan + beginner difficulty, and it generates a few beginner-friendly dishes to work through progressively.

Scenario 3: Holiday gatherings Hosting guests and need to plan a full meal. Specify headcount, taste preferences, and budget range, and it plans a complete menu including cold and hot dish balance.

Scenario 4: Health management Currently cutting calories and need high-protein, low-calorie meals. Specify those constraints and every generated recipe stays within range, with nutrition analysis to confirm.

Quick start

# Clone the repo
git clone https://github.com/liu-ziting/what-to-eat
cd what-to-eat

# Install dependencies
npm install

# Configure API key
cp .env.example .env
# Edit .env and add your OpenAI API key

# Start dev server
npm run dev

# Open browser at http://localhost:5173

One-click deploy: Supports Vercel and Netlify with ready-to-use deploy buttons. Or deploy to your own server with a custom API endpoint.

Tech stack:

• Vue 3.4 + TypeScript 5.3+
• Tailwind CSS 3.4+
• Vite 5.0+
• Any OpenAI-standard API

The good and the bad

What I loved:

• Recipe quality is surprisingly good — detailed and actually actionable
• Comprehensive cuisine coverage, especially strong for Chinese food
• Nutrition analysis and wine pairing are genuinely useful
• Dynamic AI provider switching is flexible, choose models as needed
• Modern Vue 3 + TS stack, decent code quality
• Vercel one-click deploy, up and running in 5 minutes
• Clean UI without unnecessary bells and whistles

What frustrated me:

• You must provide your own AI API key, which can get expensive without free credits
• AI food images sometimes look very different from reality — don’t take them too seriously
• Food fortune-telling is more entertainment than practical
• Recipe generation speed depends on API response, sometimes takes a few seconds
• Project is still early, some features (like favorites) are basic
• No mobile app, browser experience on phones is mediocre
• No specific open-source license listed

Bottom line

what-to-eat solves a very specific problem: using AI to reduce the decision cost of cooking. 3.3k stars shows plenty of people have this need.

It’s not a professional recipe app (like 下厨房/Xiachufang) — it’s an AI-assisted cooking companion. Its value is: when you don’t know what to eat, it gives you inspiration; when you know what you want but don’t know how to make it, it gives you detailed steps.

For people who enjoy cooking but don’t want to stress over “what to eat” every day, this project is worth trying. Deploy your own instance with your own API key, and you’ve basically got a personal AI chef.

What I find most interesting: the author calls it a “vibe coding” project. And honestly, the whole experience is exactly that — a relaxed vibe where AI handles the thinking and you just cook. No overthinking, input your needs, get results, start cooking.

For Vue developers, the codebase is also worth studying — Vue 3 + TS + Tailwind, clean project structure, and the dynamic config system implementation is quite elegant.

---

About the Author

Liudingyu is a full-stack developer and heavy GitHub user. With 900+ starred repos over the past 3 years, this site only covers tools I’ve actually used or deeply researched.

📧 Found a great tool to recommend? Email [email protected]