GuidesAPI EndpointsAPI Changelog
Live AI Demo
API Changelog
Live AI Demo
Back to All

Parsing API now keeps accents and line breaks in HTML text

7 days ago by Rajaa El Hamdani

POST /text/parsing now reads texts that contain HTML tags or HTML entities without losing their accents, line breaks, list items or words in angle brackets. This applies to both parsing models, quicksilver and atlas; plain text without tags or entities is parsed exactly as before.

Action required

  • Now: none. POST /text/parsing keeps its parameters (texts, parsing_model, output_object), their defaults and its response format.
  • Later: none. Nothing in this update is deprecated.

What changed

  • HTML entities: accented letters written as HTML entities (such as é) keep their accent in the parsed text and in the extracted values. Before, pièces came back as pieces.
  • Line breaks: plain text that contains an HTML entity keeps its line breaks, instead of being merged into one paragraph.
  • HTML layout: HTML input keeps one paragraph per line.
  • Lists: list items in HTML input come back as - item, without a stray backslash (\- item).
  • Angle brackets: words in angle brackets that are not HTML tags, such as <freelance ou CDI>, are kept as text instead of being dropped.
  • Uppercase tags: HTML tags written in capitals (such as <P>) are recognised and converted like lowercase tags.

Performance

Job offers sent with HTMLBeforeNowChange
Entity lists (F1, exact match)0.460.484% better F1

Measured on our internal benchmark of 100 job offers, on the values found in the offers that contain HTML tags or entities, on 2026-09-30. Offers in plain text score the same as before.

Example

curl -X POST https://api.hrflow.ai/v1/text/parsing \
  -H "X-API-KEY: <your_api_key>" \
  -H "X-USER-EMAIL: <your_email>" \
  -H "Content-Type: application/json" \
  -d '{
    "texts": ["<p>Heading an enterprise data science division requires strong skills in NLP and ML</p>"],
    "parsing_model": "quicksilver",
    "output_object": "text"
  }'

The response keeps the format documented in the reference (excerpt):

{
  "code": 200,
  "message": "Text extraction finished in 0.01 seconds.",
  "data": [
    {
      "parsing": {
        "skills_hard": ["NLP", "ML"],
        "tasks": ["Heading an enterprise data science division requires strong skills"]
      }
    }
  ]
}

Links