I'm a product manager who wanted a macro tracker that didn't cost £80/year. Every app I tried locked basic features behind a paywall. So I built my own, and I did it with an AI agent as my co-builder.
The result is Chowdown, a completely free AI-powered macro tracker. No subscriptions, no premium tiers, no ads. Here's how it came together.
The Problem
Tracking macros shouldn't cost money. Yet every popular app follows the same playbook: free tier with crippled features, then £7-15/month to unlock what you actually need. Barcode scanning? Premium. Food scanning? Premium. Detailed reports? Premium.
I wanted to build the opposite. Everything free, forever, funded by voluntary donations.
The Stack
- Frontend: React + Vite, deployed as a Progressive Web App
- Backend: Firebase (auth, Firestore, hosting)
- AI Food Recognition: Google Gemini API. Snap a photo, Gemini identifies the food and estimates macros
- Landing Page: Astro (static site generator), deployed on Vercel
- Analytics: Plausible (privacy-friendly, no cookies)
- AI Partner: An LLM-based agent called Rook, running via OpenClaw
How AI Food Scanning Works
The core feature is simple: take a photo of your food, get a macro breakdown. Under the hood, we send the image to Google's Gemini API with a carefully crafted prompt that asks it to:
- Identify what food is in the image
- Estimate portion sizes
- Calculate protein, carbs, fats, fibre, and calories
- Rate its own confidence level
The AI responds conversationally, breaking down each component. Users can then ask follow-up questions or adjust values before logging the entry.
Is it perfectly accurate? No. AI estimates are typically within 15-25% accuracy. But research consistently shows that consistent tracking matters more than perfect accuracy for achieving nutrition goals.
Building with AI Agents
Here's where it gets interesting. I didn't build Chowdown alone. The app itself was built using Claude Code, Anthropic's CLI tool for coding with Claude. The marketing, SEO, landing page, and growth side is handled by a separate AI agent called Rook, running via OpenClaw.
Two different AI tools, two different jobs:
Claude Code (the app):
- Building the React frontend
- Firebase backend logic
- Gemini API integration
- UI/UX implementation
Rook/OpenClaw (growth and marketing):
- Writing and deploying blog posts for SEO
- Setting up analytics and tracking
- Configuring A/B testing frameworks
- Generating marketing copy
- Managing deployments
- Researching competitors and keywords
Me (product and decisions):
- Product direction and design
- User experience and testing
- Account creation and manual submissions
- Final review of everything public-facing
It's genuinely collaborative. Rook will draft 10 SEO blog posts while I sleep, and I'll review them in the morning. Claude Code builds features in minutes that would take me hours. I make the product calls.
Lessons from the Partnership
Give clear context, not just instructions. "Write a blog post about protein" produces generic content. "Write a blog post targeting people searching 'how much protein do I need per day', positioning Chowdown as a free tool they can use to track it" produces something useful.
Review everything that goes public. The AI is fast but not infallible. It occasionally makes claims that aren't quite right, uses the wrong tone, or writes things that sound too "AI-generated." Human review catches these.
Let it handle the tedious stuff. SEO meta tags, JSON-LD schemas, sitemap generation, analytics setup. These are perfect AI tasks. Repetitive, well-documented, and easy to verify.
Don't let it make product decisions. The AI can suggest features and analyse data, but the human needs to decide what the product should be and who it's for.
The Landing Page
The landing page at chowdown.me was built with Astro, a static site generator that ships zero JavaScript by default. It includes:
- Full SEO setup (meta tags, Open Graph, Twitter Cards, JSON-LD schemas)
- A blog with 10 articles targeting nutrition-related search terms
- A macro calculator tool
- Plausible analytics with custom event tracking
- An A/B testing framework for continuous optimisation
Rook built most of the technical SEO infrastructure. I focused on the messaging and design.
Why Free Forever?
This is the question everyone asks. The answer: nutrition tracking should be accessible to everyone.
Most apps charge because they can. People are willing to pay for health tools. But "willing to pay" and "should have to pay" are different things.
Chowdown is funded by voluntary donations through Buy Me a Coffee. If people find it valuable, they can support it. If they can't or don't want to pay, they still get the full experience.
Will this model scale? Honestly, I don't know yet. But I'd rather build something genuinely useful and figure out sustainability later than lock features behind a paywall from day one.
Try It
If you're tracking macros or thinking about starting, give Chowdown a go. It works on any device with a browser; no app store download needed.
And if you're a developer thinking about building with an AI agent as a partner, I'd genuinely recommend trying it. The productivity multiplier is real, especially for solo builders.
Happy to answer questions in the comments.
Top comments (0)