pdfblah

A precise, non-destructive Swiss army knife for PDFs, from your terminal or a local app in your browser. Replace, redact, and remove text, scrub or anonymize personal data, edit metadata, watermark, stamp, number, split, merge, rotate, crop, encrypt, render, compare, fill forms, convert to and from Word, and OCR scans. Fonts, spacing, and alignment stay perfect, and nothing you didn't ask for is touched.

Most tools "edit" a PDF by painting a box over the old text and drawing new text on top, which leaves the original underneath (copy and paste still reveals it) and often adds a watermark. pdfblah rewrites the real text in the content stream, so:

• the old text is genuinely gone (pdftotext, Ctrl-F, and copy show only the new value)
• no overlay, no watermark
• your metadata (dates, Producer, XMP) is kept byte for byte, unless you choose to edit it
• alignment is auto-detected and kept, so right-aligned numbers stay flush
• fonts it cannot reproduce are refused instead of garbled

Pure Python; the core needs no system tools.

Install

One line installs the CLI and the local app:

# macOS / Linux
curl -fsSL https://pdfblah.com/install.sh | sh

# Windows (PowerShell)
irm https://pdfblah.com/install.ps1 | iex

Or straight from PyPI:

pipx install pdfblah      # recommended, isolated; or:  pip install pdfblah
uv tool install pdfblah   # or with uv

Python 3.9+, nothing to compile or sign.

Run it in your browser

Prefer a web interface? One command opens the app in your browser, fully local, with no upload, no watermark, and no account:

pdfblah gui

Drop in a PDF, add your rules, and save the edited file to your computer. Nothing leaves your machine. Or keep going for the command line.

Use

Replace the first match:

pdfblah in.pdf out.pdf --find "Old Name" --replace "New Name"

Options:

--scope all         change every match           (default: first)
--scope 3           change the 3rd match
--ci                ignore case
--word              whole word only ("cat" will not match "category")
--regex             treat --find as a regex (\1 backrefs work in --replace)
--page 2            only page 2
--replace ""        delete the text

Many rules from a file (FIND | REPLACE | FLAGS per line):

pdfblah in.pdf out.pdf --rules rules.txt

# rules.txt
Old Company Name | New Company Name | all
CONFIDENTIAL DRAFT | FINAL | ci
Jane Doe | John Smith | all word
Total | Sum | 2
delete this phrase |

Commands

Presets on the same engine: locate the real text, rewrite it.

redact removes the matched text for real (gone from pdftotext, Ctrl-F and copy) and draws a bar over each spot. --no-bar removes the text with no mark.

pdfblah redact in.pdf out.pdf --find "Account 12345"
pdfblah redact in.pdf out.pdf --find "\d{3}-\d{2}-\d{4}" --regex   # every SSN

scrub finds structured personal data (email, IBAN, credit card, SSN, phone) and removes it, or masks it. Cards and IBANs are checksum-validated, so ordinary numbers are left alone.

pdfblah scrub in.pdf out.pdf
pdfblah scrub in.pdf out.pdf --types email,credit_card --mask "[redacted]"

anonymize replaces detected data with realistic, shape-preserving fakes so a document is safe to share. The same value maps to the same fake; --seed makes it reproducible. Names are swapped only when you list them.

pdfblah anonymize in.pdf out.pdf --names "Alison Cohen,Matthew Reider" --seed 7

merge fills a template once per data row: every {{column}} placeholder becomes that row's value, one output PDF per row.

pdfblah merge template.pdf people.csv --out ./letters --name-col name

meta reports everything metadata-ish (DocInfo, XMP, pages, version, encryption), which often reveals more than you expect (author, software, timestamps). With an output file it can strip or set fields. Every other command keeps metadata intact.

pdfblah meta in.pdf                                  # report what's in there
pdfblah meta in.pdf clean.pdf --strip                # remove all metadata
pdfblah meta in.pdf out.pdf --set author="Jane Roe"  # set a field

Metadata edits can also ride along with any other command, or live in a rules file:

pdfblah redact in.pdf out.pdf --find "Acme Corp" --strip-metadata
pdfblah in.pdf out.pdf --find OLD --replace NEW --set-metadata author="Ops"

# rules.txt
@strip-metadata
@set-metadata author = Redacted Dept
CONFIDENTIAL | PUBLIC | all

Toolkit

Beyond editing text, pdfblah does the page-level and document-level jobs you usually reach for several tools to do. All keep your metadata intact.

Pages

pdfblah combine a.pdf b.pdf c.pdf -o all.pdf     # concatenate
pdfblah combine *.pdf -o binder.pdf --toc --tabs # a meeting binder: clickable Contents
                                                 # page, bookmarks, numbered edge tabs
                                                 # (--titles "A,B,C" --toc-font serif)
pdfblah split in.pdf -o parts/ --every 1         # or --ranges 1-3 4-6
pdfblah pages in.pdf out.pdf --keep 3,1,2        # keep / reorder (or --drop 4)
pdfblah rotate in.pdf out.pdf --degrees 90 --pages 1-2
pdfblah crop in.pdf out.pdf --margins 20,20,20,20

Marks (honest overlays, drawn on top; your text underneath is untouched)

pdfblah watermark in.pdf out.pdf --text DRAFT --tile --opacity 0.2 --rotation 45
pdfblah stamp in.pdf out.pdf --image logo.png --position top-right
pdfblah number in.pdf out.pdf --format "Page {n} of {total}"
pdfblah bates in.pdf out.pdf --prefix ACME --digits 6 --start 1

Security & size

pdfblah protect in.pdf out.pdf --password secret --no-copy   # AES-256 + permissions
pdfblah unlock in.pdf out.pdf --password secret
pdfblah optimize in.pdf out.pdf --downsample-dpi 150         # shrink the file
pdfblah attachments in.pdf -o out.pdf --add report.csv       # list/add/extract files

Render, extract, inspect

pdfblah render in.pdf -o images/ --dpi 150 --format png
pdfblah extract in.pdf --text                # or --images -o out/
pdfblah form in.pdf --list                   # or --fill data.json out.pdf [--flatten]
pdfblah compare a.pdf b.pdf --visual --out-dir diff/
pdfblah signatures in.pdf --validate         # read/validate (needs pdfblah[sign])

Clean scans (white paper, crisp ink — made for sheet music, works on any scan; needs the render extra: pip install "pdfblah[app]")

pdfblah clean scan.pdf out.pdf                   # background -> pure #FFFFFF
pdfblah clean a.pdf b.pdf c.pdf -o cleaned/      # batch: one cleaned PDF per input
pdfblah clean scan.pdf out.pdf --strength strong # dark or stained pages
pdfblah clean scan.pdf out.pdf --bilevel         # pure 1-bit black & white

Estimates each page's background (paper tone, lighting, stains), divides it out, and remaps levels so paper becomes exactly white while ink keeps its anti-aliased edges. --dpi sets output resolution (default 300). A 10-page scan takes a few seconds. Output pages are re-rendered images; run ocr after if you want selectable text. On a Mac, curl -fsSL https://pdfblah.com/clean-scan-mac.sh | sh installs a Finder Quick Action: select PDFs, right-click, Quick Actions > Clean Scan.

The pipeline (steps joined by |, in the words the CLI already speaks)

pdfblah do 'tidy | rotate auto | clean | ocr | shrink 200kb' scan.pdf -o clean.pdf
pdfblah do 'dark' book.pdf -o book-dark.pdf         # sugar: dark, sepia, ink navy,
pdfblah do 'watermark DRAFT | sanitize' in/ -o out/ # shrink, straighten
pdfblah do @discovery.recipe evidence/ -o stamped/  # a recipe file IS a pipeline

pdfblah help            # the map: every command, grouped
pdfblah help compress   # one command's manual, with a worked example
pdfblah verbs           # one line each; | marks pipeline verbs
pdfblah completions zsh # shell tab-completion (bash too)

Fit under a size cap, search everything, batch it all

pdfblah compress in.pdf -o out.pdf --target 200kb   # hard cap: quality ladder until it fits,
                                                    # text stays selectable; honest report on a miss
pdfblah find "invoice 4471" ~/docs -r               # grep for PDFs: file, page, snippet
pdfblah extract in.pdf --tables -o out.csv          # the tables, out as CSV
pdfblah repair broken.pdf -o fixed.pdf              # rebuild a file that won't open
pdfblah rotate in.pdf out.pdf --auto                # fix upside-down/sideways pages (lossless)
pdfblah pages in.pdf out.pdf --insert notes.pdf --every 2   # a notes page after every 2 pages

pdfblah batch discovery.recipe evidence/ -o stamped/   # one recipe over a folder; in the recipe,
                                                       # 'bates --start auto' numbers ACROSS files

Share safely, split smart, re-ink, verify

pdfblah sanitize in.pdf -o safe.pdf              # strip metadata, comments, JS, attachments,
                                                 # old revisions; --dry-run shows what it carries
pdfblah split in.pdf -o parts/ --at "Invoice #(\S+)" --name "{1}.pdf"   # split by content,
                                                                         # named from the page
pdfblah split book.pdf --spread pages.pdf        # cut two-page scans apart (--order rl)
pdfblah recolor in.pdf -o dark.pdf               # dark mode baked in; sepia, or ink=navy
pdfblah links in.pdf                             # every link + bookmark checked (--offline)
pdfblah form t.pdf --fill-from rows.csv -o out/ --name "{Employee}.pdf"  # one PDF per row

Tidy bloated PDFs (drop blank pages and exact repeats; needs the render extra: pip install "pdfblah[app]")

pdfblah tidy in.pdf --dry-run                # preview: which pages would go, and why
pdfblah tidy in.pdf -o out.pdf               # drop blanks + pixel-identical repeats
pdfblah tidy in.pdf -o out.pdf --keep-blank  # only dedupe (or --keep-duplicates)

Deterministic by design: a page counts as blank only when it renders with almost no ink AND has no extractable text (a lone "Page 4" footer survives), and as a duplicate only when its rendered pixels exactly match an earlier page. The report names every dropped page and the reason. No AI, nothing uploaded.

Check accessibility basics (the honest 60 seconds)

pdfblah access report.pdf     # tagged? language? alt text? title? text layer?
                              # failing = not accessible; passing != certified (veraPDF's job)

Archive it (PDF/A; needs Ghostscript, pdfblah doctor checks)

pdfblah pdfa in.pdf -o archive.pdf            # PDF/A-2b: fonts embedded, sRGB output
                                              # intent, XMP id; --level 1|2|3

Convert and OCR (need system tools; run pdfblah doctor to check/install them)

pip install "pdfblah[convert]"               # PDF <-> Word (also needs LibreOffice for -> PDF)
pdfblah convert in.pdf  -o out.docx          # PDF to editable Word
pdfblah convert in.docx -o out.pdf           # office (DOCX/ODT/PPTX/XLSX) to PDF

pip install "pdfblah[ocr]"                   # searchable scans (needs Tesseract + Ghostscript)
pdfblah ocr scan.pdf -o searchable.pdf --lang eng      # add a real text layer to a scan
pdfblah ocr --langs                          # OCR languages available on this machine
pdfblah ocr --get-lang deu,fra               # languages are modular: grab any of them as
                                             # one file each (~/.pdfblah/tessdata), no
                                             # package manager needed; --best for the
                                             # larger high-accuracy models
pdfblah ocr scan.pdf -o out.pdf --lang eng+deu         # then mix them freely

pdfblah doctor                               # check the system tools; --install to fetch them

Library

from pdfblah import process, redact, scrub, anonymize, merge, apply_rules

process("in.pdf", "out.pdf", "999.00", "42.00", scope="all", ci=True)
process("in.pdf", "out.pdf", r"\d{4}-\d{4}", "REDACTED", scope="all", regex=True)
redact("in.pdf", "out.pdf", "Account 12345")
scrub("in.pdf", "out.pdf", types=["email", "credit_card"])
anonymize("in.pdf", "out.pdf", names=["Alison Cohen"], seed=7)
merge("template.pdf", [{"name": "Alice"}], "./out")

Each call returns a report dict (ok, count, refused, reason, ...).

What it does not do

Scanned PDFs (image only, no text layer) can't be edited directly; run pdfblah ocr first to add a real text layer. Fonts that can't be reproduced are refused rather than rendered wrong; the app can substitute a similar standard font, but only when you ask, and the report records it.

Guides

Step by step, with pictures, at pdfblah.com/guides:

• Find and replace text in a PDF
• Redact a PDF (really remove the text)
• Remove personal data (PII) from a PDF
• View, strip, or edit PDF metadata
• Bulk find and replace in a PDF
• Run pdfblah on your own machine

Two ways to use it

• On your machine (free, this package): the command line, or the local app in your browser with pdfblah gui. Nothing is uploaded.
• Online at pdfblah.com: nothing to install, handy for a quick one-off. Upload, preview free, download.

License

MIT, (c) 2026 Kuvop LLC.