AI INDEXING

You decide what the AI learns. And you see every status.

NexDok extracts text from PDF, Word, CSV, Markdown, HTML, JSON and XML, splits it into passages and creates embeddings for Search and Ask. Nothing is indexed until someone asks for it or publishes the version.

Try it ↓Talk to us

REAL STATUS FOR EVERY VERSION…Indexing✓IndexedTText only!Failed + reason

WHAT GOES IN

Nine formats the AI can read. And the file is always kept.

Formats with text get indexed. Excel and PowerPoint are stored and downloadable, but never sent to the AI.

PDFPDF● Indexable

Text by page

pypdf

DOCWord .docx● Indexable

Paragraphs and tables

python-docx

CSVCSV · TXT · MD● Indexable

Plain text

UTF-8

WEBHTML · JSON · XML● Indexable

Includes RDF/XML

UTF-8

XLSExcel · PowerPoint● Stored only

XLSX, PPTX, DOC, XLS, PPT

AI: Not supported

IMGScans● No OCR

PDF without a text layer

Text only

Stays in NexDok: the original file — always downloadable, even if indexing fails.

Reaches the AI: text passages from the version you chose to index.

HOW IT’S INDEXED

From a file to passages the AI can find.

1 · EXTRACTThe text

PDF page by page; Word with paragraphs and tables; the rest as text.

2 · SPLIT512 words

Passages of 512 words that overlap by 64 so ideas aren’t cut.

3 · EMBEDOne vector per passage

Each passage gets a vector that represents its meaning.

4 · READYFor Search and Ask

The version shows “Indexed” and its content can now answer.

You give the order: with “Index for Ask” on the version, or by publishing it. Uploading a file doesn’t index it.

TRY IT — SAMPLE WORKSPACE

Index three files. See what NexDok says about each.

“Acme Legal” · version v2 of each document.

About PDFv2

Index for Ask

AI: INDEXING…

Extracting text and creating passages.

AI: INDEXED

15 of 15 passages embedded. Available in Search and Ask.

This document is indexed and ready for AI search and Q&A.

Sample data.

STATUSES THAT DON’T MISLEAD

Every version says where it stands, and why.

01Per version

Each version has its own status and its own error reason.

02Exact reason

Password, damaged PDF or no text: the error says so.

03Partial coverage

If some vectors are missing, the version is still searchable.

04Recovery

Interrupted indexing is marked as failed after 30 minutes.

What you see on each version

The badges in the library and version history.

BadgeMeansSearch and AskAction

AI: IndexedPassages and vectors readyYes—

AI: Indexing…Queued or processingNot yetWait

AI: Text onlyText without vectorsNoRetry indexing

AI: FailedError with a reasonNoRetry indexing

HONEST BY DESIGN

The AI learns only what you decide.

Opt-in

Uploading doesn’t index. You decide — or it happens on publish.

Drafts stay out

Indexing a draft never replaces the published version in Search and Ask.

No OCR

Scans without text aren’t indexed — and it says so.

Doesn’t use your AI quota

Indexing doesn’t count as an AI query.

PUBLISH = INDEX

Publish the version and NexDok indexes it.

When an admin publishes a version, NexDok queues it for indexing. If it’s already indexed, it’s left as is. If indexing fails, the publication still stands.

PublishIndex for AskRetry indexingStatus per version

Remote Work Policy · v2AI: Indexed

PASSAGES15of 15

WORDS6,40018 pages

OVERLAP64words

SearchAnalyze (×4)Ask

Banner: This document is indexed and ready for AI search and Q&A.

WHEN SOMETHING FAILS

The reason — not a generic error.

A version that can’t be indexed explains why, keeps the file and lets you retry.

Protected_Contract.pdf · v3

AI: FAILED

Indexing failed: pdf_password_required. The document is kept — use “Retry indexing” on the version below.

FILE

Downloadable · 2.1 MB

Every version with its status

Formats with text

PDF, Word, CSV, TXT, Markdown, HTML, JSON, XML.

Up to 25 MB

Per file.

Any member

Can request “Index for Ask”; publishing indexes automatically.

No add-ons

Included in every workspace.

What should your AI be able to answer?

We’ll help you decide which documents to index, and when.