Sokole1
commented
11 months ago See #131 which is a PR to allow for .pdf in QA: Active File. Although the changes are not complete until there is proper handling of large files. EDIT: PR now contains changes to allow for applying all the various prompts to text selected outside of Markdown such as PDF and the Copilot Chat itself.
Things to add:
- [ ] Allow users to select a PDF page range for large files
- [x] Allow the use of selected text inside PDF as context (not implementing the parsing of PDF annotations like highlights or comments for now)
- [ ] Support more file formats
- [ ] Allow using folders as context
Dev Notes:
- Document loaders appear to work only in Node.js, so I couldn't use PDFLoader. The docs were a bit misleading as the "Node-only" label is only attached to the "File Loaders" section under "Integrations". https://js.langchain.com/docs/modules/data_connection/document_loaders/how_to/pdf. It would be nice if there was a way to use these, or else a custom parser will be needed for each new file type.
- Uses PDFJs loaded in with https://docs.obsidian.md/Reference/TypeScript+API/loadPdfJs (you must use this PDFJs loader or else npm installing pdfjs-dist clashes with Obsidian's installation)
- Text extraction outside of PDF uses
activeWindow.getSelection()?.toString();. The activeWindow appears to unselect the text when the command prompt is open, so set up keyboard shortcuts to get around this.
As mentioned in #49, #117. Be able to select regions/pages of a PDF for chat context. Ideally, be able to handle figures, images, math, etc. inside the PDF.
I will explore the implementation of this feature and note down any progress under this issue. Help is welcome.