I vibe-coded a native macOS app in a weekend to escape bloated web wrappers.
Hey Product Hunt! 👋
I’m the maker of Poirot, a new companion app for Claude Code that is launching tomorrow. I wanted to share the slightly meta story of how it came to life over a single weekend.
If you use Claude Code, you know the CLI is incredibly powerful, but reading raw JSONL transcripts to track the AI's tool invocations and reasoning is a painful experience. I wanted a GUI to visualize this workflow, which led to a fun, meta thought: 'I wonder if I can build a companion app for Claude Code... using Claude Code'.
The Weekend 'Vibe-Coding' Sprint
I decided to treat it as an experiment. Using Claude Code as my co-pilot, I vibe-coded the entire application in a single creative burst. We built the JSONL parser, the SwiftUI interface, the architecture, and even the hand-written mock tests in just one weekend. Claude Code effectively helped me build the exact tool needed to investigate its own behavior!
Escaping the Bloated Web Wrapper
When building AI dev tools right now, the default path is usually an Electron app or a heavy web wrapper. I specifically wanted to push back against that. I wanted Poirot to feel like a first-class citizen on your Mac.
Instead of web tech, Poirot is built strictly native using pure SwiftUI and Swift 6 with (also strict) concurrency.
It allowed me to release the app with a tiny footprint, the entire app is under 6 MB; it's completely private: It works 100% offline with zero login, tracking, or analytics; deeply Integrated with a spotlight-style ⌘K fuzzy search, native keyboard shortcuts, and SF Symbols with native bounce and pulse animations that feel right at home on macOS.
I’m really curious to hear from the maker and developer community here:
1. Have you tried vibe-coding an entire architecture in a single weekend burst with an AI co-pilot?
2. As a user, do you still actively seek out strictly native apps (under 10MB, offline, fast), or do you feel web wrappers have gotten 'good enough' for daily dev tools?
I'd love to hear your thoughts, and if you want to see the result of this weekend sprint, the code is completely open-source.



Replies