Local Office extraction at no token cost

Index your Word, Excel, and PowerPoint without spending AI tokens

AisDrive opens docx, xlsx, and pptx files on its own server and extracts the text right there: zero calls to a model, zero tokens consumed. Only PDFs and images require AI vision.

Most AI managers pay a model to read each document you upload, even if that document is a Word file with plain text inside. AisDrive does not: what can be read locally is read locally.

AisDrive classifies each file you send to index into three extraction paths before touching anything. Text and markup formats (txt, md, csv, tsv, json, xml, html, htm, and svg) are decoded directly in UTF-8, removing markup when necessary. Office OOXML (docx, xlsx, and pptx) is opened like the zip file it truly is, and the text is extracted using local libraries. Only the third path —PDFs and images (pdf, jpg, jpeg, png, webp, gif)— goes through a multimodal model, because there is no text to read, just pages to view.

The work of the office path is done by the indexing worker, not the AI. A .docx is processed with mammoth, which preserves paragraphs, lists, and tables instead of returning a jumbled XML. A .pptx is traversed slide by slide, in true numerical order (the 2 before the 10), and each one is marked in the text so you know which slide a response comes from. A .xlsx is read by crossing the shared strings of the workbook with the cells of each sheet, so amounts and dates that appear as literal numbers are not lost along the way: each row comes out as a line of values.

This changes the calculation of what it costs to have your drive indexed. In text and Office formats, extraction is logged at zero input tokens and zero output tokens: the only consumption left in indexing those files is the embedding, the vector that enables semantic search and chat. In a PDF or a screenshot, however, you also pay for extraction via vision. If you upload entire folders of work documents, the difference shows up invoice by invoice.

And what cannot be extracted does not enter the queue: AisDrive checks the extension before consuming quota or tokens and rejects the file with a clear warning, instead of leaving you with an indexing doomed to fail. The limit per file is 20 MB. AisDrive is a mini-app of the AIsToHire platform; this functionality is explained at https://aistohire.com/aisdrive-file-manager-116/f/extract-text-office-without-tokens

Benefits of Local Office extraction at no token cost

Office at zero cost

docx, xlsx, and pptx are extracted on the server itself, and extraction is logged at 0 input tokens and 0 output tokens.

Text and markup also free

txt, md, csv, tsv, json, xml, html, htm, and svg are decoded locally, cleaning up the markup when present.

Formatted Word document

the .docx is processed with mammoth, preserving paragraphs, lists, and tables instead of flattening everything.

Excel without losing figures

the .xlsx is read by combining the shared strings from the workbook with the cells of each sheet, so literal numbers and dates are included in the index.

PowerPoint slide by slide

the .pptx is traversed in actual numerical order, and each slide is identified within the extracted text.

AI only where needed

the vision model is reserved for PDF and images, the only case where there is no text to read.

No external calls on the local route

the office suite does not leave the infrastructure to be read by a third-party model.

No doomed queues

if the extension is not extractable, AisDrive informs you immediately and does not consume any quota or tokens.

Sophie Blake

Upload your Office documents and stop paying to read them

Create your drive, send a Word, Excel, or presentation to index, and check what it consumes: the embedding and nothing more.

Extract text from Word, Excel, and PowerPoint without spending AI tokens | AisDrive