First, wanted to say the new beta is working really well for me, there’s lots of changes and improvements all over the place. I’m still working out new capabilities, and it’s nice to have seen the discussions around the role of AI and how that’s fed into a private, well-considered, and flexible implementation - it’s not just AI glop for AI glop’s sake! I’m very excited to spend more time with the automatic tag suggestions.
So my not-quite-bug not-quite-feature-request is to have a means to trigger the AI panel in settings to refresh available models. If I flip running models in local Ollama, the only way to notify DEVONthink the models have changed is to quit and restart it - the UI in settings does not pick up changes to a running server, and closing/reopening the preferences panel has no effect either. Maybe just something that refreshes the models available when entering/exiting the AI preferences panel?
(Sorry that my first comment about DT4 isn’t in the licensing thread, but I think the team needs a break after a release, not the most trafficed thread I’ve seen here!)
Exactly! Too many people think either (1) it’s all about AI (which it’s not) or (2) nothing has really changed since the interface isn’t drastically different. Thank you for saying that.
I’m finding that features I used to look at, and wished I had a use for (or thought my data didn’t fit) suddenly make a lot more sense.
I\m very pleased to see the Graph inspector chewing between - I’m assuming - DOI embedded PDFs, with that lovely graph I first found use of in DEVONagent. I have some requests† there but I’ll wait until after the launch party is over.
And as others have said - the AI integration is about as optimal as I can imagine. I respect the planning, thought and conceptualisation that went into it.
And pleased that there’s now a permanent internal name for scripting! I can update the excellent LaunchBar integration scripts which have been the only thing that’s failed on me (and that’s because it was using the DT3 references in its scripting)
† oh okay then, I’d like to be able to drag the nodes around to improve legibility - most of the documents it finds are prefixed with author names, so the display tends to show document icons and connections, but discovering the actual files requires opening them, rather than hovering or clicking.
