Skip to content

Reading files and PDFs

When you connect a file source — Google Drive, OneDrive, or Dropbox — CorpusIQ doesn't just see filenames. It reads the text inside your documents and answers from it. Ask "what's the deadline in the vendor contract in my Drive?" and CorpusIQ opens the PDF, pulls the text, and answers — citing the file.

This page explains how that text extraction works, and where its limits are.

Which files it reads

Across Google Drive, OneDrive, and Dropbox, CorpusIQ reads the common document formats:

  • PDF — the main one, and the one with the most care behind it (below).
  • Word (.docx, and older .doc where possible).
  • Excel (.xlsx) — sheets and rows.
  • PowerPoint (.pptx) — slide text.

The same engine runs no matter which of the three file connectors the document lives in. Connect Dropbox or Drive or OneDrive — the reading behaves the same way.

How PDF reading works

PDFs are deceptively hard. A clean-looking invoice can store its text in any order, split across columns, or locked inside form fields. CorpusIQ uses a three-tier strategy and automatically picks the best one that works for each file:

  1. Form-field reading first. Many business PDFs — tax forms, e-filed paperwork, structured applications — store their real values in named form fields rather than free-flowing text. CorpusIQ reads those fields directly. This is the most reliable data in the document, so it's surfaced first.

  2. Layout-aware extraction. For everything else, CorpusIQ reads the text using each character's actual position on the page. That's the part that keeps tables and multi-column pages intact — a label stays matched to its value instead of getting scrambled the way a plain text dump would. A two-column report or a line-item table comes out readable, not jumbled.

  3. Plain-text fallback. If the layout-aware pass isn't available for a given file, a straightforward text reader runs as a safety net, so you still get an answer.

You can also point CorpusIQ at a specific page range in a long document, or have it search the document for a term and return only the pages that mention it — useful when the file is 200 pages and you care about one clause.

The one limit worth knowing

CorpusIQ reads PDFs that contain real, digital text — the kind produced when a document is exported or saved as a PDF. That covers the vast majority of business documents.

What it does not do today is read a PDF that's purely a scanned image — a photo or scan of a page with no embedded text behind it. For those, CorpusIQ will tell you honestly that there's no extractable text rather than guess. Reading text off scanned images is a separate capability (optical character recognition), and it isn't part of the file connectors today.

If a scanned document doesn't read, the quickest fix is usually to open it in the source app and re-save or export it as a text-based PDF, which most tools can do.

What this unlocks

Because CorpusIQ reads the contents and not just the file list, you can ask questions that span your documents the same way you ask about your sales or marketing data:

  • "Summarize the statement of work in the contract in my Drive."
  • "What payment terms are in the latest invoice in Dropbox?"
  • "Find the renewal date in the lease PDF on OneDrive."

The answer comes back with the source file cited through a read-only retrieval tool. Direct MCP does not retain raw customer files or full connector response payloads; operational logs and optional indexed search follow their published lifecycles. See privacy-and-security.md for what that means.