pdfblah 0.2.0

Real find and replace on the actual text in a PDF. No overlay, metadata preserved.

pdfblah

Real find and replace on the actual text in a PDF, from the command line.

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
• the original metadata (dates, Producer, XMP) is preserved byte for byte
• alignment is auto-detected and kept, so right-aligned numbers stay flush
• fonts it cannot reproduce are refused instead of garbled

Pure Python (pdfplumber + pikepdf). No system dependencies.

Install

pipx install pdfblah      # recommended, isolated; or:  pip install pdfblah

On a Mac with Homebrew, use Homebrew's pipx:

brew install pipx && pipx install pdfblah

Also works with uv: uv tool install pdfblah.

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")
--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 |

Library

from pdfblah import process, apply_rules, parse_rules_file

process("in.pdf", "out.pdf", "999.00", "42.00", scope="all", ci=True)

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

What it does not do

Scanned PDFs (image only, no text layer) cannot be edited. Fonts that are not embedded and not standard, or use a custom encoding, are refused rather than rendered wrong. This is by design: a wrong-looking edit is worse than a clear "no".

Hosted version

Want it without installing anything, or for a non-technical colleague? The hosted version at pdfblah.com does the same edit in the browser: upload, preview for free, download.

License

MIT, (c) 2026 Kuvop LLC.