I’ve read the blog post and the built-in help, and I’m still rather at a loss. Why would I want to use DEVONthink’s built-in AI rather than go to the Web and, if necessary, upload documents to our chatbot of choice?
One obvious answer is that using DEVONthink eliminates the need to upload anything, and that would indeed be a benefit. And I tried it just now, asking a question of a relatively short PDF. The progress indicator spun for a few minutes and then I got an error message as follows:
search the web, Wikipedia or PubMed on its own, even when using local models like Gemma 3
download links (e.g. of web pages or even item links) to retrieve additional information
optionally change properties (e.g. tags) and contents
process annotations of PDF, RTF(D) and Markdown documents, e.g. summarize them
return item links (e.g. links to sections/pages in Markdown, EPUB or PDF documents), e.g. in a summary or table depending on your prompt
use multiple models from various providers in one chat
handle also third-party document formats that were indexed using Spotlight’s plug-ins
In addition, contrary to uploading documents in the chat app of your choice, DEVONthink performs various steps to increase the security & privacy. See DT4 - privacy when using AI - #6 by cgrunenberg
And of course it’s much more convenient than switching apps and uploading documents
Briefly: Some time ago, I entered my credit card number into ChatGPT to use its API with applications.
Some time later, I entered my ChatGPT API key into DEVONthink.
This week, when I tried this query, I was confused about the error about rate-limiting, because I was sure I had set a rate limit of $20 or $25/mo or so. I had not done any queries previously. And the ChatGPT billing page was showing my usage as zero.
After doing research into how ChatGPT API billing works, I discovered that it’s not enough to give them a credit card number. A user must also “charge” their account with a certain amount. It’s like buying tokens at an arcade or using a prepaid phone — you have to give your money to OpenAI before you can use the service.
Incidentally, I also changed the credit card I was using. That had nothing to do with my troubleshooting; my wife and I no longer want to use that particular credit card for my business expenses.
As is often the case when troubleshooting tech problems, I did not take notes as I went, and I cannot guarantee which step or steps fixed the problem.
(I mentioned my lackadaisical record-keeping to friends who are professional software developers, along with one who previously had a long career working fix-break for a corporate IT department. "I’m sure that you, professionals at this kind of thing, keep detailed records of your troubleshooting attempts. They looked embarrassed and said, "Ummmm… " So I feel better about my own lackadaisical habits — and now I have this record, which will hopefully help me and others should this problem recur.)
