Skip to content

Feature

Fill, redact, split and compress PDFs

Not every document starts as a template. Formfeed also works on the PDFs you already have: it fills a form by the names of its fields, removes names and account numbers from the file itself and checks that they are gone, cuts a PDF into parts, makes it smaller without losing anything, and draws its pages as pictures. Every tool takes a PDF you upload, which is processed and not kept, or one stored in Formfeed, and answers like a render, with a link to the result.

Free plan with 100 units a month, no credit card

Redaction that removes, then checks

POST /v1/pdf/redact takes the texts to remove, found on every page whatever their spacing, and rectangles of the page as a viewer shows it. What lies there is taken out of the file, not painted over: the text, the pixels of an image under a rectangle, a link or a comment it touches. Then the result is read again, with its metadata, bookmarks, attachments and form, and while a text can still be found the call fails with 422 redaction_incomplete and delivers nothing (what is checked).

  • Found whatever the spacing. An IBAN written in groups of four is found by its characters run together. Letters are compared after Unicode normalisation and, unless match_case is set, regardless of case; whole_words finds a text only as a word.
  • Rectangles in points. x and y from the top-left corner of the page as a viewer shows it, its rotation applied. 72 points are an inch.
  • A box, another colour or none. A black box goes over each removed place; fill names another colour, and null leaves the place blank.
  • Never stored. The texts you name are neither stored nor logged.
The top of the example non-disclosure agreement’s first page, with both parties named
Before: the top of the NDA from the examples.
The same part of the page with the second party’s name and address removed, in the running header too
After: “Olvarest Ltd” and “2 Example Road, Exampleville” removed wherever they stood, the header included.
TerminalShell
curl https://api.formfeed.dev/v1/pdf/redact \
  -H "Authorization: Bearer $FORMFEED_API_KEY" \
  -F file=@nda.pdf \
  -F 'text=["Olvarest Ltd","2 Example Road, Exampleville"]'

Fill a form by the names of its fields

POST /v1/pdf/info lists the fields of a form with their full names, labels, types and options, free on every plan: here applicant.name, a text field labelled “Full name”, and membership, a radio group of basic, plus, team. POST /v1/pdf/fill takes values by those names: text, true or false for a check box, an option of a choice or a radio group. The form stays fillable, or with flatten the values are drawn into the pages and the fields removed (fill).

  • Strict, or lenient. A name the form lacks, or a value its field does not take, fails with 422 naming the field and never the value. With strict: false both are left out and listed in the render’s form.skipped.
  • What a viewer would run. No calculation or display format of the form runs; the render’s form.warnings names the fields that have one.
  • Forms that cannot be filled. A form that needs a password, a signed form and a form only XFA viewers fill are refused, each with its own code; a hybrid form is filled through its AcroForm, its XFA part removed.
  • Never stored. The values are neither stored nor logged; the render keeps the names of the fields it filled.
An empty membership application form with nine fields
Before: a form as an office sends it, with 9 fields.
The same form filled in: name, address, a radio button, a date from a list and a ticked check box
After: filled by name and flattened.
TerminalShell
curl https://api.formfeed.dev/v1/pdf/fill \
  -H "Authorization: Bearer $FORMFEED_API_KEY" \
  -F file=@membership-application.pdf \
  -F 'fields={
    "applicant.name": "Jane Doe",
    "membership": "plus",
    "start": "1 February 2027",
    "newsletter": true
  }' \
  -F flatten=true

Split into parts

POST /v1/pdf/split cuts a PDF every N pages, or into one part per page selection such as 1-3,5 or 4- for page 4 to the last, and answers one ZIP of the parts named after the source: report.zip holds report-1.pdf and report-2.pdf. A split that yields one part answers that PDF. A PDF/A is refused, because its parts would carry a claim nobody checked (split).

The first page of report-1.pdf
report-1.pdf, 1 page.
The first page of report-2.pdf
report-2.pdf, 1 page.
TerminalShell
curl https://api.formfeed.dev/v1/pdf/split \
  -H "Authorization: Bearer $FORMFEED_API_KEY" \
  -F file=@report.pdf \
  -F every=1

Pages as pictures

POST /v1/pdf/to-image draws pages as PNG or JPEG at 36 to 300 dpi, 150 when you name none: one page answers the picture, several a ZIP of them, named by page. One call draws up to 100 pages, and each picture costs a unit, as an image render does (to-image).

The first page of the example invoice as a picture
invoice-1.png: 1242 × 1756 pixels at 150 dpi, 131 kB.
TerminalShell
curl https://api.formfeed.dev/v1/pdf/to-image \
  -H "Authorization: Bearer $FORMFEED_API_KEY" \
  -F file=@invoice.pdf \
  -F pages=1 \
  -F dpi=150

Smaller, without losing anything

POST /v1/pdf/compress rewrites a PDF without changing what it shows: objects packed into compressed streams, identical streams stored once, images recompressed losslessly and fonts cut down to the glyphs the document uses. Image quality is never lowered, and a file the rewrite cannot make smaller comes back as it was. Five examples of this site, measured on 18 November 2026, came out 15% to 42% smaller (compress):

ExampleBeforeAfterSaved
price-list.pdf272 kB225 kB17%
nda.pdf147 kB124 kB15%
us-invoice.pdf141 kB81 kB42%
mahnung.pdf114 kB70 kB38%
invoice.pdf90 kB60 kB33%
TerminalShell
curl https://api.formfeed.dev/v1/pdf/compress \
  -H "Authorization: Bearer $FORMFEED_API_KEY" \
  -F file=@price-list.pdf

A PDF you upload, or one stored in Formfeed

Every tool takes the PDF as multipart/form-data with the file as file, up to your plan’s output size and 100 MB, processed and not kept; or as JSON naming a stored PDF in source: the ID of a render, such as the invoice your template just made, or a PDF of the file library by its name. The result is a render like any other, with its download link, kept as long as your plan keeps files (sources).

JSONJSON
POST /v1/pdf/split
{
  "source": "rnd_7da09ab6cd9946b798c187c2a9350852",
  "ranges": ["1", "2-"]
}

Merge, watermark and password in the render itself

A render’s post block appends further PDFs after the document, stamps a text or image watermark and encrypts the result with AES-256, and the same three run on stored PDFs through /v1/pdf/merge, /v1/pdf/watermark and /v1/pdf/protect. The examples show a draft watermark and a payslip with a password.

From code, the command line or an agent

  • SDKs. client.pdf.redact(), fill(), split(), compress(), toImage() and info() in TypeScript, the same in Python with to_image().
  • Command line. npx formfeed pdf redact nda.pdf --text "Olvarest Ltd", and fill, split, compress, to-image and info beside it.
  • AI agents. The MCP server’s redact_pdf, fill_pdf_form, split_pdf, compress_pdf, pdf_to_image and get_pdf_info (AI agents).
  • No-code workflows. Steps for the PDF tools in the no-code apps, platform by platform (integrations).

Plans and units

The tools are part of the Starter plan and above. Each costs 0.5 units on a live key, drawing pages one unit per picture, and /pdf/info is free on every plan (pricing).

FAQ

Is redacted text really gone?
Yes. It is taken out of the file, not painted over, and the result is read again before it is delivered: while any of the texts can still be found, on a page, in the metadata, a bookmark, an attachment or a form field, the call fails and no file is delivered.
Can I fill a form I have never seen?
Ask /pdf/info first: it lists every field with its full name, label, type and options, and those names are what /pdf/fill takes.
Are uploaded files kept?
No. An upload is processed and not kept; only the result is stored, like any render, for as long as your plan keeps files.

From the blog

  • Redacting a PDF so the text is really gone →

    A black box over a name hides it from the eye and from nobody else. Three times the hidden text came back in public, what a redaction has to remove besides the words on the page, and the one letter our own tool left behind while we wrote this.

More features

Start rendering today

The free plan includes 100 units a month and free test renders. Change plans when you need more.