File Formats

Falara supports 9 file formats across 8 format families. Each format has a dedicated processor that extracts translatable text, preserves structure and formatting, and reassembles the translated output.

---

Supported Formats

Format	Extensions	Description
HTML	.html, .htm	Markup is extracted and replaced with typed placeholders; reassembled after translation
Markdown	.md, .markdown	Document structure (headings, lists, code blocks) is preserved; only prose is translated
Word	.docx	Paragraphs and table cells are extracted; character-level formatting (bold, italic) is preserved
Excel	.xlsx	Individual cells are translated; formulas and cell references are skipped
PowerPoint	.pptx	Text boxes and speaker notes are extracted and translated
XLIFF 1.2	.xlf, .xliff	Standard exchange format; <trans-unit> source/target pairs are translated
XLIFF 2.0	.xlf, .xliff	Newer XLIFF standard; <segment> elements are translated
Eurotext-XLIFF	.xlf	Proprietary Eurotext format with extended metadata and CDATA-embedded HTML
JSON	.json	Key-value pairs; values are translated, keys and structure are preserved
PDF	.pdf	Text is extracted from the document and translated; original layout is not reconstructed

---

Limits

Limit	Value
Single file upload	50 MB
Batch total	50 MB

---

How File Processing Works

1. Upload — File is received and MIME type is validated against the declared extension
2. Extraction — Format processor extracts translatable segments; non-translatable content (code, formulas, metadata) is preserved as-is
3. Translation — Segments pass through the multi-agent pipeline
4. Injection — Translated segments are written back into the original document structure
5. Download — Reassembled file is served via GET /v1/jobs/{job_id}/download

The injection step happens on-demand at download time, not during the pipeline. The original file structure is never permanently modified.

---

XLIFF Handling

For XLIFF files, the processor reads <source> elements and writes translations back into <target> elements. Existing <target> content is replaced.

Eurotext-XLIFF files preserve: - CDATA wrappers on text content - HTML markup embedded inside CDATA (block tags, attribute order) - Proprietary metadata attributes

---

JSON Handling

For JSON files, only string values are translated. Keys, numbers, booleans, and structural elements remain unchanged. Nested objects and arrays are traversed recursively.

{
  "title": "Willkommen",
  "items": [
    { "label": "Startseite", "url": "/home" }
  ]
}

In this example, "Willkommen", "Startseite" are translated. "title", "items", "label", "url", and "/home" are preserved.

---

PDF Handling

For PDF files, text content is extracted and translated. Note that the original PDF layout is not reconstructed — the translated output is delivered as extracted text segments.

---

Format Detection

Format is detected from the file extension, then validated against the actual MIME type of the uploaded file. A mismatch returns 415 Unsupported Media Type.