Chomptron

AI-powered recipe generator that transforms ingredients into delicious recipes using Google Gemini AI. Deployed on Google Cloud Run at chomptron.com.

What It Does

Enter ingredients you have in your kitchen, and Chomptron generates creative, practical recipes complete with measurements, instructions, cooking time, and serving sizes.

Related MCP server: Mealie MCP Server

Local Development

npm install
export GEMINI_API_KEY="your-api-key-here"
export GEMINI_MODEL="gemini-3.1-flash-lite"  # Optional, defaults to gemini-3.1-flash-lite
npm start

Visit http://localhost:8080

Get API key: https://makersuite.google.com/app/apikey

Environment Variables

• GEMINI_API_KEY (required) - Your Google Gemini API key

• GEMINI_MODEL (optional) - Model to use, defaults to gemini-3.1-flash-lite

• PORT (optional) - Server port, defaults to 8080

Model Configuration

The Gemini model can be configured via the GEMINI_MODEL environment variable:

Recommended Models (as of December 2025):

• gemini-3.1-flash-lite (default) - Best free tier limits: 15 RPM, 1,000 RPD

• gemini-2.5-flash - 10 RPM, 250 RPD

• gemini-2.0-flash - 10 RPM, 200 RPD (⚠️ unstable quota, often shows limit: 0)

• gemini-1.5-flash - Legacy model, may have better limits than 2.0

December 2025 Quota Shift:

Google overhauled free tier quotas in December 2025:

• gemini-2.0-flash was removed from fully unauthenticated free tier

• Many accounts see limit: 0 errors for newer models without billing enabled

• Free tier quotas don't automatically reset monthly

Fixing "Limit: 0" Errors:

If you're seeing quota errors with limit: 0:

1. Switch to gemini-2.5-flash-lite - Best free tier model currently available

2. Enable billing (Pay-As-You-Go) - Linking a credit card (even if you don't spend) moves you from "Limited Free" to "Tier 1" and unlocks promised free quotas

3. Check your region - EEA, UK, and Switzerland have restricted free tier access

4. Monitor usage - Visit /api/usage endpoint or https://ai.dev/usage

Free Tier Limits (as of Dec 2025):

Testing

npm test

Tests validate:

• Modular File Structure: Verification of external styles.css and app.js linking

• CI/CD & DevOps: Docker and Cloud Build configurations

• API Integrity: Health check endpoints (/health, /ready) and recipe generation

• Feature Robustness: Recipe history, scaling logic, and favorites

API Endpoints

• GET /health - Liveness check, returns service status

• GET /ready - Readiness check, verifies AI configuration and shows current model

• GET /api/usage - Usage statistics (total requests, quota errors, model info)

• POST /api/generate-recipe - Main recipe generation endpoint

  • Request Body:

    {
      "ingredients": "chicken, tomatoes, garlic",
      "dietaryPreferences": {
        "vegan": false,
        "vegetarian": false,
        "glutenFree": false,
        "dairyFree": false,
        "nutFree": false,
        "shellfishFree": false,
        "eggFree": false,
        "soyFree": false
      }
    }

  • Response:

    {
      "success": true,
      "recipe": "**Recipe Name:** ...",
      "cached": false
    }

• POST /mcp — MCP Streamable HTTP. Tool generate_recipe (ingredients + optional dietary flags). Unauthenticated. Daily cap MCP_DAILY_RECIPE_LIMIT (default 20), persisted to MCP_BUDGET_FILE so a process restart on the same instance does not reset it.

Architecture

Chomptron is built as a serverless application on Google Cloud Run for cost efficiency and automatic scaling.

Tech Stack:

• Backend: Node.js 24 + Express

• AI: Google Gemini (configurable model, defaults to gemini-3.1-flash-lite)

• Frontend: Modular Vanilla HTML/CSS/JavaScript (Clean separation of concerns)

• Storage: Browser localStorage for recipe history

• Platform: Google Cloud Run (serverless)

• CI/CD: Cloud Build

• Domain: chomptron.com

Why Serverless?

• Scales to zero when idle → $0 cost (vs. $5-50/month traditional hosting)

• Auto-scales from 0 to 1000+ instances based on traffic

• Zero maintenance - no servers to manage, patch, or configure

• Perfect for AI workloads - handles burst traffic and CPU-intensive recipe generation efficiently

Performance Optimizations:

• In-Memory Caching: Per-instance recipe cache (24-hour TTL, max 100 recipes)

• Structured Recipe Parsing: Extracts recipe components for better display and scaling

• Client-Side Rate Limiting: Prevents excessive API calls

• Smart Retry Logic: Handles quota errors gracefully with exponential backoff

Usage Guide

Generating Recipes

1. Enter your ingredients in the text area

2. (Optional) Select dietary preferences/allergies

3. Click "Generate Recipe ✨"

4. View your structured recipe with:

   • Recipe name and metadata

   • Scaled ingredients list

   • Step-by-step instructions

   • Cooking tips

Recipe Features

• Scale Servings: Use +/- buttons to adjust serving size (0.25x to 4x)

• Rate Recipes: Click stars to rate recipes (1-5 stars)

• Add Notes: Type personal notes in the notes field

• Favorite: Click the star button to favorite recipes

• Print: Click print button for print-friendly view

• Share: Click share button to generate shareable URL

Recipe History

• Click the 📚 button (top-right) to open recipe history

• Search recipes by name, ingredients, or content

• Filter by favorites

• Export all recipes as JSON or text

• Click any recipe to reload it

Dark Mode

• Click the 🌙 button (top-left) to toggle dark/light mode

• Preference is saved automatically

Monitoring

Health checks:

curl https://chomptron.com/health
curl https://chomptron.com/ready
curl https://chomptron.com/api/usage  # View usage statistics

View logs:

# Recent logs
gcloud run logs read chomptron --region us-central1 --limit 50

# Live stream
gcloud run logs tail chomptron --region us-central1

Console dashboards:

• Cloud Run Service

• Cloud Build History

• Artifact Registry

Features

Core Functionality

• ✨ AI-powered recipe generation using Google Gemini

• 🍳 Creative recipe names and instructions

• 📏 Precise measurements and serving sizes

• ⏱️ Cooking time estimates (prep, cook, total time)

• 🎨 Modern Visual Polish: Glassmorphism design with staggered entrance animations

• 🔠 Premium Typography: Outfit and Inter fonts for a contemporary feel

• ⚡ Serverless, auto-scaling infrastructure on Google Cloud Run

Recipe Management

• 📚 Recipe history with localStorage persistence (up to 100 recipes)

• ⭐ Favorites system to mark and filter beloved recipes

• 🔍 Search and filter through saved recipes

• 💾 Export recipes to JSON or text format

• 📋 Quick access to past recipes via sidebar panel

• 📝 Add personal notes to each recipe

• ⭐ Rate recipes with 5-star rating system

Recipe Customization

• 🥗 Dietary Preferences & Allergies: Vegan, Vegetarian, Gluten-Free, Dairy-Free, Nut-Free, Shellfish-Free, Egg-Free, Soy-Free

• 📊 Recipe Scaling: Adjust serving sizes from 0.25x to 4x with automatic ingredient scaling

• 🖨️ Print-Friendly View: Clean print layout optimized for printing recipes

• 🔗 Shareable URLs: Generate shareable links for recipes (Web Share API support)

Recipe Display

• 📋 Structured Recipe Format: Parsed display with organized sections:

  • Recipe name

  • Serving size and timing information

  • Ingredients list

  • Step-by-step instructions

  • Cooking tips

• 🎨 Dark/Light Mode: Toggle between themes with persistent preference

Performance & Optimization

• 💾 Recipe Caching: In-memory cache reduces API calls for similar ingredient combinations

• 🔄 Smart Retry Logic: Automatic retry with exponential backoff for quota errors

• ⚡ Rate Limiting: Client-side rate limiting prevents excessive API calls

Technical Features

• 🔍 Health monitoring and readiness checks

• 🔎 SEO optimized with meta tags, Open Graph, Twitter Cards, and structured data

• 📱 PWA support with manifest.json

• 🤖 robots.txt and sitemap.xml for search engine indexing

Recipe Data Structure

Recipes are stored in browser localStorage with the following structure:

{
  "id": "timestamp",
  "ingredients": "chicken, tomatoes, garlic",
  "recipe": "Full recipe text...",
  "recipeName": "Extracted recipe name",
  "timestamp": "2025-01-01T00:00:00.000Z",
  "favorite": false,
  "rating": 0,
  "notes": "",
  "servingSize": 4,
  "dietaryPreferences": {
    "vegan": false,
    "vegetarian": false,
    "glutenFree": false,
    "dairyFree": false,
    "nutFree": false,
    "shellfishFree": false,
    "eggFree": false,
    "soyFree": false
  }
}

Caching Strategy

The application uses in-memory caching on the server side:

• Cache Key: Normalized ingredients + dietary preferences

• TTL: 24 hours

• Max Size: 100 recipes per instance

• Scope: Per-instance (serverless instances are ephemeral)

• Benefits: Reduces API calls for similar requests hitting the same instance

SEO Features

Optimized for search engines and social sharing with meta tags, Open Graph, Twitter Cards, structured data (JSON-LD), sitemap, robots.txt, and PWA support.

Buildkite Pipeline

This repo includes a Buildkite pipeline at .buildkite/pipeline.yml that runs on a self-hosted GCP agent provisioned by buildkite-gcp-agent.

Pipeline structure

Push / PR
  └── Three steps dispatched in parallel:
        ├── :prettier: Format check   (npm run format -- --check)
        ├── :eslint:   Lint           (npm run lint)
        └── :nodejs:   Tests          (npm test)
  └── Annotate — consolidated pass/fail surfaced in the Buildkite UI

The key difference from the GitHub Actions workflow, which runs format → lint → test sequentially in a single job: these three steps have no dependencies on each other and run simultaneously. On a pool of agents this distributes the work; even on a single agent the explicit dependency graph documents intent and makes the pipeline trivially scalable.

The deploy stage is intentionally omitted — Cloud Run deployments are handled by the existing GHA workflow which holds the production GCP credentials. CI (fast feedback) and CD (production access) are kept in separate systems as a security boundary.

Pipeline setup

Connect the repo via Buildkite → New Pipeline → point at github.com/swantron/chomptron. Buildkite reads .buildkite/pipeline.yml automatically on each push.

Agent targeting

All steps run on the gcp queue:

agents:
  queue: gcp

Contributing

This is a personal project, but suggestions and improvements are welcome!

License

MIT