All tests run on an 8-year-old MacBook Air. All results from shipping 7 Mac apps as a solo developer. No sponsored opinion.
I've shipped 7 paid Mac apps. Monthly AI spend: $0. Here's the exact stack, what each tool does, and where the limits are.
The Stack at a Glance
| Layer | Tool | Cost |
|---|---|---|
| Code autocomplete | Ollama + qwen2.5-coder:1.5b + Continue | $0 |
| Chat / debugging | Claude free + Gemini free | $0 |
| In-app AI features | Gemini API free tier | $0 |
| Complex reasoning | Claude free (targeted use) | $0 |
Total: $0/month.
Layer 1: Local Autocomplete
Ollama + qwen2.5-coder:1.5b + Continue extension in VS Code.
Runs locally. No API key. No quota. Suggestions in under a second even on old Intel hardware.
This covers the majority of what I type: boilerplate, repetitive patterns, obvious method completions. It's not GPT-4. It doesn't need to be. It handles the low-value completions so I can use cloud AI budget on the high-value questions.
Layer 2: Chat and Debugging
Claude free tier for hard problems. Gemini free for quick questions.
The rule: if a wrong answer would cost me 30+ minutes, I use Claude. If a wrong answer is obvious immediately, Gemini is fast enough.
In practice:
- Rust borrow checker fights → Claude
- "What's the flag for this adb command" → Gemini
Both have daily limits. I rarely hit them because I'm selective about what I ask.
Layer 3: In-App AI Features
Gemini API free tier.
Two shipped apps use Gemini API for user-facing features. The free tier has been sufficient for both. Flash is fast enough for real-time use cases. The quota hasn't been a production issue.
For in-app AI, local LLM isn't viable — I can't require users to have Ollama installed. Gemini API free tier is the only practical $0 option.
Where the Free Stack Breaks Down
Long debugging sessions. Claude free tier has daily limits. A 4-hour session on a hard problem will exhaust them.
Complex reasoning at volume. If you're making dozens of substantial AI requests per day, free tiers aren't enough.
Production AI features at scale. Gemini free tier has rate limits. If your app goes viral and 1000 users hit the API simultaneously, you'll need a paid plan.
The Upgrade Trigger
When app revenue exceeds the monthly cost of the AI tier above, upgrade.
Until then, the free stack ships real products. I'll upgrade when the free limits become the bottleneck. They haven't been yet.
If this was useful, a ❤️ helps more than you'd think!
👉 HiyokoBar → https://hiyokomtp.lemonsqueezy.com/checkout/buy/f9b85321-6878-40aa-a472-ff748d6de2d5
X → @hiyoyok