Why we built GoSosh

I started building GoSosh because the signal got harder to follow at the exact moment AI was supposed to make it easier.

In 2023, every reading app added a chatbot. In 2024, every news app added AI-generated summaries. By 2025, “AI-powered” had stopped meaning anything — every app said it, most of them just meant “we send your data to OpenAI and paraphrase what comes back.” The summaries got the facts wrong often enough that I’d end up reading the original anyway, which meant the AI layer was costing me time, not saving it.

What I actually wanted from a tracking app was the opposite of a chatbot. I wanted to see the pattern across sources — when five different outlets are covering the same event, who’s actually being talked about across the day, what’s gaining momentum before it breaks. Not just for news. For the stocks I follow. For the people whose work I track. For the niches I care about. The work of reading is the point. I just wanted help deciding which fifteen minutes of reading were the right fifteen minutes.

That’s a job AI is genuinely good at, when you point it at the right problem. Not “write this for me” but “tell me which of these items are about the same thing.” Not “summarize the conversation” but “identify which people and companies are showing up across sources.” Classification. Clustering. The boring AI tasks that nobody puts in a launch announcement.

Apple shipped Foundation Models in iOS 18.1 — a small language model that runs entirely on the phone. Fast enough for live use, never sends data anywhere. It’s exactly the right scale for the work I wanted to do. I started building.

The privacy story fell out of the architecture. If the AI runs on-device, you don’t need a server. If there’s no server, there’s no account. If there’s no account, there’s no email, no profile, no tracking-history database somewhere with my name on it. I didn’t set out to build a privacy app. I set out to build a signal-tracking app, and chose the tools that worked. The privacy was a side effect of the tools.

I think about that side effect a lot now. Every other “AI-powered” reading product I’ve tried sees what I track. Most of them log it. Some of them train on it. None of them have to — the work doesn’t require any of that data leaving my phone. The cloud-LLM architecture is convenient for the company building the app, not for the person using it.

GoSosh is the version I wanted. Small AI doing specific work. No chatbot. No summaries. No data leaving the device. The original sources, one tap away, in a clean reader. For any topic I want to track.

If that’s the version you wanted too, you’ll know within ten minutes of using it.