Nadgar

2026-07-13

Nadgar is an app that had been on my mind for some time.
I use AI a lot, and it often helps that it’s right at hand.
My main subscription is ChatGPT, although earlier I used apps based on the BYOK approach a lot, bring your own key. In that kind of app, you paste in a key, for example from the OpenAI API, and then you’re billed based on token usage. This works (or used to work) if you use AI only a little and your API token usage ends up cheaper than paying $20 for a normal subscription. For me personally, paying for tokens in the API is no longer worth it, because I use a lot of apps based on AI agents, like Codex, and they eat quite a lot of tokens. So I prefer paying for a full subscription, because it’s more cost-effective.

Still, the sentiment for BYOK apps remained, so when creating Nadgar, I based it on exactly that approach.

The main idea behind Nadgar is the ability to use AI from the Apple Watch. AW is a little computer worn on your wrist, and I was missing a good app that would let me use its capabilities this way, while also not making me buy a separate subscription. I first started thinking about something like this around two years ago, and back then the idea was pretty simple, a regular app with an AI chat, like ChatGPT. In the meantime, though, new concepts and possibilities appeared, and the initial implementation is already more powerful. First of all, in Nadgar we have one session, so that in such a small app we don’t also need a complex UI for managing multiple conversations. Secondly, the app can be connected to many providers and custom endpoints, which makes it possible, for example, to do TTS and STT with an OpenAI key, but send messages to a Hermes agent hosted on your own server. This lets us communicate with a model that has far more tools than a regular question-answering chat.