You could make it an optional premium feature. My guess is that most people, especially premium users, would not mind parting with 300mb of disk space in exchange for the value that embeddings would bring.
Sure, we could do that. Perhaps you have to download the model yourself before using the feature.
My guess is that Apple will add this. If they don’t add it next year, we’ll take stock.
The LLM stuff is all such a moving target, that we want to be sure we aren’t leaving aside other projects just to do something that will be obsolete in 6 months.
That’s why we are generally following Apple’s lead. If they put something in, it will usually be supported for many years. Hopefully next year they add a powerful LLM with the Private Cloud Compute.
Personally, I am using LLMs a lot, but don’t use them in Agenda anyway. I tend to use them via ChatGPT or Claude Code, so integration in Agenda makes no sense. I haven’t even used our Writing Tools support much.
The ability to semantically search seems like a feature that would be powerful. We are taking a small step with that in iOS/macOS 26, but I expect that to get more powerful over time, and if embeddings are added, we can easily slot it in there.
1 Like
The challenge with Apple including an embedding model is another one of those challenges that you face as a platform developer. You have to make trade offs, and I suspect that given the fact that the choice of embedding model is something that has long-term ramifications, e.g., you can’t easily switch from one to another without having to re-process everything over again, they chose to leave that up to developers.
I totally understand not wanting to add 300MB to every Agenda download - although it’s not a huge amount, it’s not something that you or users would want to have to do for every Agenda update etc. I do think, however, it’s something that could be downloaded via web once per device without negative impact.