Tilly The Coder

Tomorrow 👀

💅

@ShaneDRosenthal @nativephp Weeekend plans changed 📱

If anyone wants to see/play with the latest @nativephp Super Native - now works on Android AND iOS, would love some feedback 👇

@ShaneDRosenthal Doing yoga in the 'skia' is even more helpful!

Yoga makes you Flexible

transcription added:

Also adding support for audio transcription on audio file upload (up to a reasonable limit).

I am seeing how far I can get by building a PHP install manager for Windows using @nativephp and GPT 5.3 codex. Wanted a GUI for updating the php.ini settings especially. I am considering replacing Laravel Herd and potentially locked features. UI is basic for now.

Happy International Women's Day! 👩‍💻💃

Using GPT-1.5 realtime instead of elevenlabs, and works pretty well across languages. Adding it to web and iOS app

@nativephp microphone plugin + ElevenLabs = 🤯

It's 100% slower. Each tool call requires a new request to the LLM provider. E.g.,: Request 1: "I see a reference to User. Let me call ls." (Wait for response...) Request 2: "Okay, User.php exists. Let me call read_file('User.php')." (Wait for response...) Request 3: "Wait, User.php uses a HasRoles trait. Where is that? Let me search..." (Wait for response...) But it's also more efficient. As much as it's impractical for developers to read a 5k index.php it's faster. I worked at a company where it was legacy code with giant .php files, and it felt faster than when I went home and asked it to do some work on my own project, again purely because it requires less requests.

.@levelsio genuinely curious: for AI coding tools like Claude Code, is it easier to understand one giant index.php with thousands of lines vs a split codebase with hundreds of smaller files?

Honestly if you just want a fast answer and it's not extremely complex GPT-OSS 120b from @cerebras is the best model to use. Which is why it's the default model for my app unless user chooses another one. It's super fast and cheap.

What if my @nativephp gets a voice assistant built in? Like how you can speak with ChatGPT/Claude on iOS/Android?

yes 👀

Added haptic feedback today. Creating a @nativephp plugin is super simple, GPT-5.4 "vibe" coded it in 10m

Data-streaming is great, but there is cases where web sockets + queuing makes more sense. It's a balanced approach

The reason I started building this app, is actually because of Gemini. When I was on a train, I kept having to get Gemini to retry to generate messages as it, does not fallback to queuing when the user's connection is poor. If a user is on 4/5g and not wifi, there's a good chance the connection could drop. Thus, it makes sense to consider using queuing + web sockets for such use case, especially if you use the AI model, ask a complex question and come back 5m later to check it.

I just uninstalled Gemini, ChatGPT and Claude. Going to use my own app for all AI related questions, doing that will enable me to put it into beta soon while loving the product.

Oh, were you worried about PHP's performance on mobile? #dontworry 😏