Kasimir Docs DEEN To the app →

File formats in detail

Every supported format: what Kasimir reads from it, what the preview shows, where the limits are.

Kasimir reads far more than PDF and Office. This page lists every supported format, explains what Kasimir extracts, what the preview looks like and where the limits are. New since 16 Sept 2026: bills of quantities, e-invoices, e-mail files, CAD drawings and medical documents.

How a file is read

Every file takes the same route, whether it was uploaded, synced from the cloud or found in a live folder:

flowchart TD
  A["File arrives<br/>Upload · cloud sync · live folder"] --> B{"Determine type"}
  B -->|"MIME is specific"| C["MIME wins"]
  B -->|"MIME is generic<br/>+ extension is BoQ or CAD"| D["Extension wins"]
  C --> E["Pick reader"]
  D --> E
  E --> F["docparse-Sidecar<br/>BoQ · e-invoice · mail · CAD · FHIR/CDA"]
  E --> G["Worker reader<br/>PDF · Office · text · image"]
  F --> H{"Health data?"}
  H -->|"FHIR/CDA without release"| I["blocked<br/>preview and download locked"]
  H -->|"allowed"| J["Text form + preview"]
  G --> J
  J --> K["Search index and citations"]

Specialist formats are read by a dedicated service (the docparse sidecar), everyday formats by the application server itself. The result is always the same: a text form for search, chat and citations, plus a browser preview in most cases.

How the type is detected

Browsers and cloud providers frequently mislabel specialist formats: a .dxf drawing arrives as text/plain, an .x83 bill of quantities as application/xml, a .dwg as application/octet-stream. So the rule is: if the reported type says nothing and the extension points to a bill of quantities or a CAD drawing, the extension wins. A specific type (PDF, Office, image) always keeps priority.

What this means in practice: nothing to rename, nothing to configure. Upload the file exactly as your specialist software produced it.

Everyday formats

Format

Extensions

What Kasimir reads from it

PDF

.pdf

text per page; scanned pages additionally via text recognition (OCR)

Word, Excel, PowerPoint

.docx .xlsx .pptx .doc .xls .ppt

body text, table cells, slide content

OpenDocument

.odt .ods .odp

same as Office

Text and data

.txt .md .csv .tsv .json .xml .html .log

plain text, unchanged

Source code and config

.ts .js .py .sql .yml .sh .ps1 and many more

plain text; never executed

Images

.png .jpg .webp .heic

image description and contained text via a vision-capable model

Specialist formats

These four groups are deliberately available as documents only — via upload in the Documents area, cloud sync or a live folder. They are not meant as quick chat attachments, because they are read structurally and have lasting value in your knowledge base.

Bills of quantities (GAEB)

Twelve extensions, one format: .x81 through .x86 (GAEB DA XML) and .d81 through .d86 (GAEB 90). Whether a file is XML or the older record format is decided by its content, not its extension.

  • Kasimir reads the title structure, item numbers, short texts, quantities, units and both unit and total prices.

  • Notes appear as their own lines so preliminary remarks are not lost.

  • The preview shows the BoQ as a readable page; every item has an anchor, so a citation in chat jumps straight to the right item.

In text form an item looks like this — the citation card derives its anchor from it:

## Titel 3.2 Erdarbeiten
Pos. 3.2.40 · Baugrubenaushub loesen und laden · 120,000 m3 · EP 18,50 EUR · GP 2.220,00 EUR

Limits: GAEB 2000 (.p8x) is not supported. For GAEB 90 the item type (contingency, alternative, lump sum) is not decoded — in the XML format it appears in brackets after the item number. A DA XML file stored as UTF-16 is rejected.

E-invoices

Supported are XRechnung in both syntaxes (UBL and UN/CEFACT CII) and ZUGFeRD — the PDF with an embedded XML invoice. Invoices are detected by content, so an .xml file needs no special naming.

  • Kasimir reads the invoice header (number, date, currency), seller and buyer, service period, payment terms and payment account, all line items with quantity, unit and total price, plus the tax rates.

  • For ZUGFeRD, Kasimir reads the embedded XML invoice — the authoritative values come from the data, not from the picture of the page.

  • The preview renders the invoice as a clean page; line items carry anchors.

E-mail files

.eml (standard format) and .msg (Outlook). Kasimir reads sender, recipients, date, subject and the message body — stripped of layout for HTML mail.

  • Attachments are read as well, even when they sit in nested mail parts (multipart/related) — exactly where mail clients hide embedded images.

  • An attachment is always delivered as a file, never rendered in the browser. That is deliberate: mail is foreign content.

  • The preview shows the mail as mail — headers on top, body below, attachments as a list.

CAD drawings

.dxf and .dwg (AutoCAD R2000 to R2018). Important to understand: Kasimir does not treat a drawing as an image, it reads its content.

  • It reads all texts and labels, layer names, block attributes (e.g. title block entries), dimensions and leader texts.

  • This is how search finds room numbers, component designations or revision states written inside the drawing.

  • The preview is a rendered image of the model space on white background with layer colours.

Medical documents

HL7 FHIR (JSON resources and bundles) and HL7 CDA (clinical documents as XML). Both are detected by content, regardless of the extension.

These formats are blocked by default. Health data is specially protected under Article 9 GDPR. An administrator has to release it explicitly under Administration → Security.

Without that release Kasimir refuses processing and remembers the kind of document — preview and download stay locked afterwards, too. The gate applies to every way in: upload, chat attachment, cloud sync and live folder. If the check itself fails, the file is blocked rather than let through.

With the release in place, Kasimir reads FHIR resources with type and identifier (e.g. conditions, observations, medication) and, for CDA, the document sections. FHIR citations jump to the named resource in the preview.

Preview and citations

What happens when you open a document depends on its kind:

Kind

Preview

Citation anchor

PDF, images

directly in the browser

page number

Bill of quantities

readable BoQ page

item number

E-invoice

invoice page

line number

E-mail

mail view with attachment list

CAD

rendered drawing image

FHIR

resource list

resource (type and id)

CDA

sections

none (page opens at the top)

Office, OpenDocument

converted to PDF

page number

anything else

download

A download always returns the original, never the converted version.

Limits at a glance

  • A single file may be 25 MB.

  • Specialist formats need the docparse service. If it is unreachable, a bill of quantities stays unreadable — on purpose, because raw markup in the search index would be worse than a clear error message.

  • BoQ reading and previews have a 15-second budget, CAD 20 seconds. Very large drawings can hit that; the text form is unaffected.

  • Bills of quantities, e-mail files and CAD drawings are document formats, not chat attachments.