When I open an Epub ebook and ask the AI in the Inspectors → Chat pane to summarize this Epub ebook, the chat shows the status fetched.
Does this mean DEVONthink has uploaded the entire contents of the book to the LLM？
I checked my OpenRouter token usage and it’s about 100,000 tokens, while the book itself has around 320,000 characters.
I have no idea what DT does. But what would you expect if you asked for the summary of a book? Should only the first chapter be uploaded? The last? The marketing blurb on the cover?
DEVONthink doesn’t per se upload raw or original documents, only e.g. the plain text or parts of it when necessary. This depends also on the used model and Settings > AI > Chat > Usage.
Actually, I am ok that the “Chat in Inspectors” would upload the entire book to the LLM, so I may get better results.
So, if I want to upload the entire book to the LLM, should I opt for advanced models like Claude Sonnet 4.5 or GPT-5.1? And should I select “Best” under Settings > AI > Chat > Usage?
The Best option would be ideal for such large documents but most modern models (e.g. GPT 5 Mini or Gemini 2.5 Flash or Mistral Medium 3.x) should be able to perform simple operations like summarizing. By the way, it’s not even necessary to use the chat assistant for this usage scenario, you might want to have a look at Edit > Summarize via Chat… or Tools > Summarize Documents via Chat > … too
An upcoming release will also include the possibility to change the usage mode easily on demand.
Actually, what I need is not a simple summary. In my use case, I need to write a long prompt so that the LLM can generate a customized summary for me, and on DEVONthink’s chat page I currently do this by copying and pasting the prompt each time.
In the future, could DEVONthink’s AI support saving some commonly used prompts, so that on the chat page I could trigger these prompts with a shortcut key (similar to how Sider.ai uses the / symbol to trigger prompts)?
I agree in general that storing snippets in app is helpful for many, but if you use AI a lot, I find it more useful to use a text expander or snippet manager to keep my prompts in. That then makes it easier to use the same prompts across multiple apps.
This is planned for upcoming releases.
Thanks for your suggestion. I will try that
There’s another problem. When I open an epub document and click “edit-summarize via chat,” a chat window appears showing the document’s summary. However, when I click elsewhere, this window disappears and cannot be found again. My summary hasn’t been saved yet, so I have to click “edit-summarize via chat” again, consuming a large number of tokens again. This is very troublesome during use.
The last results of the popovers (e.g. summarizing and transforming) are cached, therefore summarizing the same document with the same options again should immediately show the summary and doesn’t use additional tokens.
