Product / Ingestion
Documents in, structure kept, ready for search.
In short
troveGEN ingestion turns files, scans and web pages into searchable passages that remember their page, section and table. It reads the document the way a person would, protects sensitive details before anything is stored, and shows what it did to every file, step by step.
What it reads
Everyday formats
PDF, Word, PowerPoint, Excel and CSV, HTML, Markdown, plain text, RTF, EPUB, OpenDocument and email (.eml).
Scans and photographs
Pages that are only pictures are read with OCR, including pages whose text is drawn as shapes. The reader picks the script by itself, so Hindi, Tamil and Arabic scans are read in their own language.
Honest quality signals
Every scan gets a text-quality rating. Obvious reading errors are repaired, and a scan that is still poor is flagged rather than quietly indexed.
Websites and feeds
Crawl a site with its linked PDF, Word and Excel files. The crawler stays on your hosts, respects robots.txt, skips logins and duplicates, and never fetches private addresses.
Paste, upload or drop a folder
Upload files, whole folders with their structure, paste text or add a link. Large uploads run in the background with progress you can watch.
What it keeps
Structure
Passages are cut along the document’s own headings and sentences. Each one carries its section path and page number, so a result can say where it came from.
Tables as tables
Tables stay whole or become one searchable entry per row. Bank-statement style layouts are rebuilt into tables so totals can be computed exactly.
Books
Chapters are found from bookmarks, headings or the contents page. Running heads and page numbers are removed, and contents and index pages stay searchable under their own names.
Figures, charts and formulas
Images are stored and linked to their passage. Optional vision models describe a figure or read the numbers from a chart into a table. Formulas are never split.
Context for each passage
Optional context lines and generated questions per passage help short or pronoun-heavy passages be found for the right reasons.
What it understands
Industry documents
Finance, healthcare, legal, education, hospitality and government documents are recognised automatically and handled the way that industry needs. Choose a type yourself and troveGEN still checks it and flags a mismatch.
18 languages verified
Hindi, Marathi, Tamil, Telugu, Bengali, Urdu, Arabic, Hebrew, Russian, Greek, Spanish, French, German, Polish, Turkish, Indonesian, Vietnamese and English, tested end to end. Right-to-left text is displayed correctly.
Duplicates and versions
Identical uploads are recognised. With versioning on, a re-uploaded document keeps the earlier version for “as of” questions, and a delete removes every version.
What it protects
Sensitive data first
Account numbers, national IDs, card numbers and similar identifiers are detected on every upload and replaced with tokens before indexing. The originals sit in a separate encrypted vault.
Classification with a receipt
Every document is labelled by type and sensitivity. Only types and counts are recorded, never the values.
What you can see and change
A timeline per document
Each document shows its stages (received, read, secured, split, embedded, stored) with timings, so a slow or failed file explains itself.
Preview before you commit
A chunking simulator shows how a sample will be split as you change settings, and a smart-setup step recommends settings from a sample.
Change your mind without re-uploading
Re-index a document with new settings, relabel its type, or move a whole project to another embedding model from the stored text.
How a file becomes searchable
Read
The right reader is chosen by a quick pre-scan of the file, with a built-in reader as a safety net.
Understand
Type, language, structure and sensitive details are identified.
Protect
Sensitive values are replaced with tokens before anything is indexed.
Split
Passages follow headings, sentences, rows and chapters, never mid-word.
Index
Passages are embedded and stored in our hosted store or your own vector database.
Questions people ask
What file types can troveGEN ingest?
PDF, Word, PowerPoint, Excel and CSV, HTML, Markdown, plain text, RTF, EPUB, OpenDocument, email (.eml), images of documents, and web pages through a crawler. Up to 100 MB per document.
Can it read scanned documents?
Yes. Scanned and photographed pages are read with OCR in the document’s own script, with a text-quality rating and automatic repair of common reading errors. A scan that cannot be read well is flagged instead of being indexed as an empty page.
How does troveGEN handle tables?
Tables are kept as tables: whole, or as one searchable entry per row with its column headings attached. Totals and averages over a table are computed exactly instead of being estimated by a language model.
Does ingestion remove sensitive data?
High-risk identifiers such as account numbers, national IDs and card numbers are replaced with tokens before indexing, on every plan. The originals are held in a separate encrypted vault and shown only to callers with explicit permission.
Can I ingest a whole website?
Yes. The crawler collects the useful pages, downloads linked PDF, Word and Excel files, and can be scheduled to re-crawl so changes are picked up.
Last updated 2026-10-26. Figures are from troveGEN’s own tests on the dates given; ask us for the method.
Load a real corpus and judge it yourself.
The free trial gives 500 pages, enough to try your own hardest documents.