Sorry for the late reply! We did this in our first MVP actually, and the result was that giving more choice over model selection resulted in worse retention / usage over time

Each models have their own behaviors / use cases (we're using Gemini now), and by opening it up to everything we needed to allow users full control over the prompts (even with defaults), which increased the learning curve significantly (and the usability of other features like Voices)

Ultimately decided to go with generous (i.e. Unlimited) paid plan

Agree. A lifetime license is the only model that makes sense for a wrapper. I wouldn't even pay $5/mo.

Totally this. I already pay for ChatGPT Plus and like it. Forcing their "own" AI for another $20/month is a nonstarter.

Can't speak to Yoink but I built a project a year or two ago that did something similar. I used a stupid simple method - input simulation (using the fantastic enigo library). And if you want context, you can simulate select all + copy, and access via clipboard. Similar if you want to replace, etc.

https://github.com/jasonjmcghee/plock

The goal of this project was to just be able to pipe / run arbitrary commands from any text context. But, that lined up well with LLMs usage.

Sorry for the late reply! Not super familiar with HN and didn't have email notifs on :/ thought my launch here flopped completely...

We use a mix of accessibility, GUI trees, and pastebins for this!

Thank you!

Yeah we received this feedback a bunch and we thought that the underlying app was different enough.. On hindsight we should probably have named it something else..