Built an Android AI agent that shows its plan first. What would make you trust it with a task?

I make Kryzz AI (free, MIT, Android). You give it a task, it shows a live checklist, does the searching and reading, and hands you a file, for example a spreadsheet comparing three phones with sources.

The thing I keep going back and forth on is trust. Right now it shows the plan first, logs every search and page it read, caps spend per task ($0.25 by default), and stores everything on the phone. Models run through your own OpenRouter key, and the APK is still a debug build. Is that enough, or would you want an approve-each-step mode?

Product page: https://www.producthunt.com/products/kryzz-ai
APK + source: https://github.com/kriddss3/kryzz-ai

Happy to give feedback on yours too, drop a link.