You might assume that our applications will do certain things automatically, such as performing OCR on imported files. However, many decisions are yours to make, not the software’s. We prefer clear opt‑in controls rather than hidden opt‑outs. Several features or actions in DEVONthink or DEVONthink To Go require setup on your part before they function on their own. These are the most common ones:
Sync: Not every database needs to be synced. You must set the same sync location on each device running DEVONthink or DEVONthink To Go, then choose which databases to sync. If you create new databases, manually enable them for syncing or import them from the sync location. For more information, have a look at our sync checklist for DEVONthink.
AI: Nothing is enabled by default, because AI always remains a private matter. If you want to use external AI, set up a connection to your provider in the AI settings and obtain any required API keys. You can also choose task-specific models, e.g., for summarization.
Automation: In DEVONthink, you must manually run automations or configure their event triggers. Scripts and batch processes must be run manually. Newly installed, created, or duplicated smart rules initially use the On Demand trigger. Add other triggers, e.g., On Import, if you want them to run automatically.
OCR: DEVONthink does not automatically OCR every image or PDF on import. Set up a smart rule to detect those documents and run OCR. To make incoming scans from a known scanning software searchable, switch on Settings > OCR > Convert Incoming Scans. In DEVONthink To Go, enable OCR > Make scans searchable for in-app scans.
In the Settings, our applications give you precise control over many options, including automatically importing email attachments, scheduling RSS updates, creating thumbnails, or transcribing media files. We believe you should decide how much an app does on its own.
After reading this blog post, I found myself a bit confused about the various OCR and transcription options in DEVONthink. I think I’ve figured it out now, but I wonder if the UI and manual could be a bit clearer?
As I understand it:
OCR as described in this post (triggered manually or by a rule) uses ABBY FineReader and is full-featured OCR which you can use to create a text layer or extract the text from a document for use elsewhere.
Settings are in Settings > OCR
The ‘Recognition’ settings under Settings > Files > Import are where I got more confused
The manual says: ‘DEVONthink can use AI to detect and save the content from certain kinds of files automatically’, and that it uses the AI model from Settings> AI > Transcription
The ‘Make text in PDF documents searchable’ option says that PDFs are ‘processed via the Vision framework’, i.e. Apple’s on-device framework, so not the AI model set under Settings> AI > Transcription (unless it does in fact use that setting for PDFs and images, in which case the manual description is slightly wrong)
‘Transcribe Text & Notes in Images’: according to the manual, ‘AI processes images’ – is this via the Vision framework again, or the model set in AI settings?
The other transcription options (audio, video, barcodes) presumably are using the AI model selected in AI settings. This wouldn’t have been obvious to me without reading the manual (I’d have assumed audio and video were using Apple’s on-device transcription). I wonder if features that use charged-for AI models ought to have some sort of AI icon next to them, or have some help text in the settings UI to explain this?
While OCR is technically the same term, we draw a sharp distinction between it and text recognition via Vision-enabled AI. Similar to Apple’s Live Text, the text is either stored in a separate document, a Finder comment, or the database’s index. There is no explicit – and portable – text layer added to a PDF document. So the comments in the blog post stand: documents are not OCR-processed just because you e.g., add a PDF document into a database.
Transcription is unrelated, as speech-to-text is an entirely separate technology.
Both recognition and transcription use the models specified in the respective AI settings.
Thanks Jim. I agree, the blog post is correct, but it sent me off on a slight tangent (which may be better posted elsewhere - please let me know if you rather I move it to a different thread).
So does the ‘Make text in PDF documents searchable’ option under Settings > Files > Import use the selected AI model, or the Vision framework (the manual says the latter, which is confusing if all the other recognition options use the selected model).
My broader point is that those recognition settings are the only settings (as far as I can tell), that relate to the selected AI model. Everything else AI-related is under the AI settings tab. I wonder if it should be called out slightly more clearly that ticking those boxes might mean that you’re paying for AI use on every import?
