A few months ago, I was building TaskFlow — a voice-based task manager app. I needed to pick an AI model for the NLP layer.

I spent an embarrassing amount of time just trying to answer one question: **which API is actually cheapest for my use case?**

The problem:

- OpenAI, Anthropic, Google, DeepSeek, Qwen — all have different pricing

- Input vs output tokens are priced differently

- Prices change constantly (dropped 80% in 2026 alone)

- Web tools exist, but none work offline or on mobile

I was manually copy-pasting numbers into a spreadsheet at midnight. That's when I thought — there has to be a better way.

So I built **AI Wallet**.

It does 4 things:

🧮 **Cost Calculator** — enter your tokens and daily volume, see real-time cost for every major model side by side

🤖 **Smart Recommender** — pick your use case (chatbot, coding, content) + priority (cheapest/balanced/best) → get top 3 models ranked with explanations powered by Claude API

💳 **Subscription Audit** — add your AI subscriptions, it detects overlap (I was paying for both ChatGPT Plus and Perplexity Pro — 70% overlap)

🔔 **Billing Tracker** — get notified 7, 3, and 1 days before any subscription charges you

Some things I learned building this:

- DeepSeek V3.2 at $0.14/1M tokens is genuinely insane value right now

- Most developers I talked to had NO idea prices had dropped this much

- The German App Store ranked it organically before I even started marketing — apparently German devs care a lot about cost optimization 🇩🇪

The app is free with a Pro tier. Built with SwiftUI, StoreKit 2, and Claude API for the recommendation engine.

Would love honest feedback — especially from anyone building with LLM APIs. What do you use to track costs right now?

🔗 App Store: https://apps.apple.com/us/app/ai-wallet-llm-cost-advisor/id6768362109

https://apps.apple.com/us/app/swipeclean-storage-cleaner/id6755654053

I didn’t have a big launch, no ads, no fancy growth hacks. Just shipped fast and kept improving based on feedback.
Here’s what worked for me:
• Started with a very narrow use case
• Focused on making the first experience smooth
• Iterated based on real user behavior (not assumptions)
• Kept the product simple instead of adding too many features