Products // 3 of 4 · Ask

Sibyl

In development

Ask your documents. Get answers you can check.

Sibyl answers questions about a page or a whole archive, finds documents by meaning, summarises long files and translates them with the right terminology. It always shows the source, and says so when the answer is not in the documents.

What goes in and what comes out

You send

  • A question, in French or English
  • One document or a whole collection

You get

  • Answers with the source highlighted
  • Search results by meaning
  • Summaries and translations

Inside Sibyl // 4 features

What it does.

Questions & answers

Ask the document. Get the cell.

Sibyl answers questions about a page, a file or a whole collection, including questions about tables and layout. Every answer cites the span, box or cell it relies on, and when the documents do not support an answer, Sibyl says so instead of guessing.

  • Grounded, or silent. Generation may use only retrieved evidence and must abstain when it is insufficient. MTRACE · arXiv 2025
  • Questions about the page. Visual question answering over layout, tables and figures, answered with the region highlighted. Cross-modal document understanding · PhD thesis 2022
  • Answers for a date. With Mnemos, Sibyl answers with the clause valid today, or on the day you ask about.

Search

Find it by meaning, by language or by look.

Sibyl searches collections by meaning rather than exact words, across phrasings and languages, and finds pages by what they look like: the same form, the same stamp, the same layout. Results come with the passage and page that matched.

  • Robust to rephrasing. Semantic query expansion with reciprocal rank fusion, evaluated on French and English. MTRACE · arXiv 2025
  • Search by image. Content-based document image retrieval with cross-modal representations. GlobalDoc · WACV 2025
  • Query spotting. Locating a query structure inside larger graphs, benchmarked on heterogeneous subgraphs. HSB-GDL · IJDAR 2026

Summaries

The page that replaces the file.

Sibyl writes short, faithful summaries of long and administrative documents, robust to OCR errors and domain vocabulary, and links each sentence back to the passages it summarises.

  • Domain-adapted summaries. Pre-training on OCR-transcribed administrative text and fine-tuning with question–answer pairs, for terminology, OCR noise and scarce labels. DocSum · WACV Workshops 2025
  • Summaries that cite. Each sentence points to its sources, so a summary can be checked rather than trusted.

Translation

Translation that keeps the terminology right.

Sibyl translates documents with the vocabulary of their domain, retrieving in-domain terminology as it goes, with compact models that can run close to the data.

See it work // illustrative

Sibyl, on fictional documents.

The demo runs in your browser on fictional documents. It shows what Sibyl returns; it is not production software and it measures nothing.

Pipeline · one document through the chain Illustrative
Supplier invoice Fictional through the chain
StageProductWhat it outputs
01 IngestGrammata · Capture & quality checks Scanned PDF, 1 page, capture quality 0.96: accepted
02 ReadGrammata · OCR & handwriting Printed French, Latin script; 3 words flagged below 0.90
03 LayoutGrammata · Layout & tables 13 regions in reading order; line-item table rebuilt as 5 rows × 4 columns
04 ClassifyKleis · Classification & routing Supplier invoice (0.97), routed to accounts payable
05 ExtractKleis · Field extraction 13 fields normalised; 1 below 0.90 held for review
06 EntitiesKleis · Names & entities 9 entities, 5 linked to known records
07 VerifyKleis · Consistency & ID checks 5 checks: 4 pass, 1 warning
08 AskSibyl · Questions & answers 3 answers with citations, 1 explicit abstention
09 SummariseSibyl · Summaries 3 sentences, each citing its regions
10 TranslateSibyl · Translation 2 regions, French → English, terminology kept
11 RememberMnemos · Document memory 4 claims written with provenance; matches a known supplier template
Illustrative: scripted outputs on fictional documents. Confidences and scores are examples, not measurements.

Try Sibyl on your documents.

We start with a pilot on a sample of your own documents, measured before anything goes to production.