Report data extraction

Reports in. Findings out.

Upload a report as a PDF, Word file or scan, forward it by e-mail or send it to the API. Parsevix returns the title, author, date, summary, key findings, recommendations and every section as structured data, in the report's own words.

  • Summary, findings and recommendations as fields
  • Every section with its title and content
  • Personal data masked automatically before the AI step

No credit card · 20 free pages every month

What you get back

Every part of a report, already defined.

The built-in Report category returns 20 fields: the report and who wrote it, the summary, conclusions and methodology, lists of findings, recommendations, sections and appendices, and the version and status. Use it as it is, or change it in your project.

Summary, findings and recommendations come back in the report's own words, and the date as an ISO date. A field the report doesn't contain, such as references, is left out, not guessed.

Report

  • titlestring
  • report_typestring
  • datedate
  • period_coveredstring

Author and client

  • authorstring
  • organizationstring
  • prepared_forstring

Summary and conclusions

  • summarystring
  • conclusionsstring
  • methodologystring
  • referencesstring

Key findings

key_findingsarray

  • findingstring

Recommendations

recommendationsarray

  • recommendationstring

Sections

sectionsarray

  • section_titlestring
  • section_contentstring

Version and status

  • versionstring
  • statusstring
  • confidentiality_levelstring

Appendices

appendicesarray

  • appendix_titlestring
  • appendix_contentstring

Your own fields

A name, a type and one sentence on what it is.

  • budget_total
  • audit_scope
  • kpi_table
  • next_review_date

Made for long documents

Twelve pages in, one result out.

Audits, research, progress and financial reports run long and put the conclusions in different places. This is what Parsevix does with them.

01

The whole report, section by section

A report of up to 30 pages goes to the AI model in one pass. Each section comes back with its title and a short summary of its content, in the report's order, next to the summary, findings and recommendations from the front pages.

  • 02

    Findings and recommendations as lists

    Each key finding and each recommendation is an item of its own, so your system can list them, count them or turn recommendations into tasks.

  • 03

    Version, status and classification

    The version, the status such as draft or final, and the confidentiality level printed on the report come back as fields, so drafts and confidential reports can be routed differently.

  • 04

    Any kind of report

    Audits, research, inspection, progress or financial reports: the report type is a field, and one category handles them all. Add fields for the figures you need from a particular kind.

  • 05

    Reports in any language

    The language is detected on every page. A report in another language fills the same fields, with ISO dates.

  • 06

    Names and addresses masked

    Names of people and companies and their addresses are masked automatically before any text goes to the AI model and put back in your result. Detection catches most personal data, not necessarily every item.

  • 07

    Every value linked to its page

    Each value is linked to the spot in the report it came from. Click it in the dashboard to see the passage on its page.

How reports get in

Upload them, forward them or send them from your system.

Dashboard

Upload reports as PDFs, Word files or scans. Check each result next to its pages, with every value marked where it was printed, and approve it.

E-mail address

Paid plans

Every project gets its own address. Forward reports to it, and each attachment becomes a document.

API

Send reports from your document management or reporting system and get the JSON in the same response, or receive results by signed webhook or through Zapier.

Request

curl https://parsevix.com/api/v1/extract \
  -H "X-API-Key: $PARSEVIX_API_KEY" \
  -F file=@energy-audit-2026.pdf \
  -F category="Report"

The request stays open until the document is read and answers with the JSON.

  • POST /documentsUpload without waiting; the result arrives by webhook or callback.
  • POST /extract-autoNo category: every field Parsevix finds, named as in the document.
  • POST /classifyOnly the document's category, for 1 credit.
Full API reference

Response · trimmed

{
"status": "Completed",
"category": "Report",
"pageCount": 12,
"language": "en",
"extracted": {
"title": "Head office energy audit",
"date": "2026-09-18",
"summary": "The head office used 412 MWh of electricity in the first half of 2026. Lighting and ventilation account for most of it, and three measures can cut use by about a fifth.",
"key_findings": [
{
"finding": "Lighting uses 38% of the site's electricity, mostly through fluorescent tubes."
},
{
"finding": "Ventilation runs at full speed around the clock, weekends included."
},
{
"finding": "Two of the five heating pumps are oversized for their load."
}
],
…
},
…
}

Security & compliance

  • PII maskedautomatically, before the AI step
  • Encryptedin transit and at rest
  • No AI trainingon your documents
  • API files deletedafter processing

Priced per page, not per feature.

Every plan reads, sorts and masks documents the same way. Picking the category is included, and a document that fails costs nothing.

See pricing
credit per page
1
free pages every month
20
a month for 500 pages
$39

Report FAQ

Reports,
answered.

What teams ask before they send their first reports.

Still have a question?

Write to us. A person reads every message.

PDF, Word (DOCX), PNG, JPG and TIFF, up to 10 MB and 30 pages each. Scanned and printed reports work too; every page is read as an image.

A file can have up to 30 pages. A longer report has to be split into files, and each part comes back as a result of its own, so fields from different parts are not merged.

It takes what the report states: its summary, findings, recommendations and conclusions. Only the section contents are summarised from each section's text. It doesn't judge whether a report is right or complete.

The built-in fields hold the report's text. For figures from a table, add a field or a list to the category in your project. Charts are not interpreted; text printed on them is read like any other text.

One credit per page: a twelve-page report uses twelve. Choosing the category is included, and if a document fails, its credits are given back.

Reports uploaded in the dashboard or sent by e-mail stay in your project until you delete them. Files sent through the API are deleted after processing unless you turn on keeping them for the project. Your reports are never used to train AI.