When a document is uploaded to Unplex, it isn't just stored — it is read, understood, and prepared so the AI can answer questions about it with precise, verifiable citations. This article explains what happens during that preparation, how long it takes, and the limits that apply.

File check: Unplex verifies the file type is supported and the file is within the size limit (see Section 3). If a check fails, a clear status is returned immediately — for example Unsupported or Too large — and the file is not processed.

PDF preview generation: regardless of the original format (Word, Excel, PowerPoint, email, etc.), a PDF version is generated so every document displays consistently and answers can highlight the exact source passage.

Text extraction: the appropriate reading method is selected automatically per document:

Indexing: extracted text is split into passages, each tagged with its exact position in the document. From this point the document can be opened, searched, and queried.

Deeper analysis (background): following indexing, Unplex generates a summary, identifies the document type, and builds a structural map (chapters, sections, tables) with layered summaries. This map allows Unplex to answer questions about long documents without losing context, and to attach a precise citation to every statement. These outputs appear shortly after the document becomes searchable.

Trustworthy answers require more than raw text. By preserving each document's structure and the exact position of every passage, Unplex can show where an answer came from — down to the highlighted sentence on the page. This is part of how Unplex prevents AI hallucinations: every claim is traceable back to the source document (see How Unplex Prevents AI Hallucinations).

The governing constraint on document size is file size, not page count: maximum 50 MB per file. Larger files are rejected with a Too large status. A document with many pages but a small file size (e.g. a digital, text-based document) is not rejected on page count alone; very large or image-heavy files reach the 50 MB ceiling before an arbitrary page count would apply. Note: for the complete file type and volume reference, see Supported File Types and Document Limits.

Ordinary digital documents: typically ready within about a minute of upload.

Scanned documents: take longer, as each page image is read via OCR — a few extra seconds per page.

Financial statements with dense tables (bank statements, portfolio valuations): the most demanding case, potentially several minutes per document.

Scanned pages: up to 30 pages of OCR per document. Beyond 30 scanned pages, any real text in the file is still indexed, but text recognition on the remaining page images is skipped and the document is marked partially processed with a warning. For scans exceeding this, splitting into parts of 30 pages or fewer is recommended.

Unplex processes up to 5 documents concurrently; additional uploads queue, with live progress and queue position shown per document. Background steps (summary, structural map) may complete slightly after the document is already searchable.

These are the standard platform limits; self-hosted and Enterprise installations can be configured differently (see Deployment Options).

Related articles: Supported File Types and Document Limits · How Unplex Prevents AI Hallucinations · How to Review Documents with Unplex