squeeze

squeeze extracts 20 kinds of data from text, including URLs, emails, IPs, hashes, and TODOs. Like grep -o, with the patterns built in.

$ cat notes.md
See [the docs](https://example.com/docs), or https://en.wikipedia.org/wiki/Squeeze_(disambiguation).
Ping ops@example.com, then check <https://status.example.com>.

$ grep -oE 'https?://[^ ]+' notes.md
https://example.com/docs),
https://en.wikipedia.org/wiki/Squeeze_(disambiguation).
https://status.example.com>.

$ squeeze --url notes.md
https://example.com/docs
https://en.wikipedia.org/wiki/Squeeze_(disambiguation)
https://status.example.com

• Recognizes values in prose, Markdown, JSON, and HTML, handling surrounding punctuation and balanced parentheses in links.
• Reads pipes, files, globs, and directory trees, respecting ignore files.
• Prints one value per line, editor locations, or JSON, YAML, and CSV.
• Available as a CLI and a Rust library.

Install

Homebrew (macOS and Linux):

brew install aymericbeaumet/tap/squeeze

mise (Linux, macOS, and Windows):

mise use -g github:aymericbeaumet/squeeze

Cargo (Rust 1.95+):

cargo install --locked squeeze-cli

All three install the squeeze command. Prebuilt binaries support Linux, macOS, and Windows on amd64 and arm64. You can also download them from GitHub Releases, verify the archive against SHA256SUMS, and put the binary on your PATH.

Shell completions: squeeze --completions <shell> prints the script for bash, zsh, fish, elvish, or PowerShell; Homebrew installs them for you.

Usage

Pick one or more finders, or none to run all of them. Results contain one matched value per line; use --with-kind to label them. With no path, or with -, squeeze reads stdin. Paths can be files, directories, or quoted glob patterns. Directory scans skip hidden entries, ignored paths, and binary files.

# Everything squeeze recognizes in a file, one value per line
squeeze app.log

# Unique URLs on a web page
curl -s https://example.com | squeeze --url --uniq

# Everyone who committed to a repository
git log --format='%an <%ae>' | squeeze --email --sort --uniq

# IPs, request IDs, and timestamps in logs, labeled by kind
kubectl logs deploy/api | squeeze --ip --uuid --datetime --with-kind

# TODOs and FIXMEs across a repository, with locations your editor understands
squeeze --todo --fixme --with-location .

# Rust sources only
squeeze --todo --fixme --with-location 'src/**/*.rs'

# Extract URLs as JSON, with match locations
squeeze --url --with-location --output json notes.md

With several finders, results come out in the order they appear:

$ echo '2026-01-15T10:30:00Z GET /health from 10.0.4.2 took 3ms' | squeeze --datetime --ip --with-kind
datetime  2026-01-15T10:30:00Z
ip        10.0.4.2

Finders

Finder	Flag	Examples
CIDR	--cidr	192.168.1.0/24, 2001:db8::/32
Codetags	--codetag, --todo, --fixme	TODO: fix this, FIXME(#42): bug
Colors	--color	#ff0000, rgb(255, 0, 0), hsl(0, 100%, 50%)
Datetimes	--datetime	2024-01-15, 2024-01-15T10:30:00Z
Domains	--domain	example.com, mail.example.co.uk
Emails	--email	user@example.com, first.last+tag@company.co.uk
Emojis	--emoji	😀, 👨‍👩‍👧‍👦, 1️⃣
Env vars	--env	$HOME, ${PATH}, ${VAR:-default}
Handles	--handle	@alice, @user@example.social
Hashes	--hash, --md5, --sha1, --sha256, --sha512	5d41402abc4b2a76b9719d911017c592
IPs	--ip, --ipv4, --ipv6	192.168.1.1, ::1, 2001:db8::1
JSON	--json	{"key": "value"}, [1, 2, 3]
JWTs	--jwt	eyJhbGciOiJIUzI1NiJ9.eyJzdWIiOiIxIn0.c2ln
MACs	--mac	00:1A:2B:3C:4D:5E, 001A.2B3C.4D5E
Modelines	--modeline	vim: set ts=4 sw=4 et:
Paths	--path	/etc/hosts, ./src/main.rs:42:10
Phones	--phone	+14155551234, (415) 555-1234
Semver	--semver	1.0.0, v2.3.1-rc.1+build.42
URIs	--uri, --url, --http, --https	https://example.com, mailto:hi@example.com
UUIDs	--uuid	550e8400-e29b-41d4-a716-446655440000

Some finders take a filter: --codetag=todo,fixme, --hash=sha256, --uri=https,ssh. --url is shorthand for the schemes people usually mean by a URL (http, https, ftp, mailto, …); --uri accepts any scheme.

Finders favor precision over recall: they recognize common formats and may skip unusual valid values. They extract candidates, without checking whether a URL is reachable or an email address exists. Use --strict for URI syntax following RFC 3986.

Output options

Flag	Description
-1, --first	stop after the first result
--last	only print the last result
--sort	sort the results
--uniq	deduplicate the results, keeping the first occurrence
--with-kind	print the finder name (uri, email, …) with each result, aligned in a terminal and tab-separated otherwise
--with-location	print path:line:column: before each result
--output <fmt>	text (default), json, yaml, csv, or none
--no-overlap	drop matches that overlap an earlier one (--precedence longest keeps the longest instead)
--copy	copy the results to the clipboard
--open	open each result with the default application
-j, --jobs <N>	scanning threads; auto (default) uses every core for directories, stdin, and files of 8 MiB and more, output of a single input kept in order

Walking a directory:

Flag	Description
--hidden	also search hidden files and directories (.git is always skipped)
--no-ignore	do not respect .gitignore, .ignore, and git exclude rules
--follow	follow symbolic links

JSON and YAML return lists of values by default; CSV returns one value per row. With --with-kind or --with-location, structured output includes kind, value, line, column, start, end, and source. Lines and columns start at 1; byte offsets start at 0 and the end is exclusive.

Matches can overlap: a URL can also contain a domain or an IP address. Use --no-overlap to suppress overlapping results. Run squeeze --help for the full list of options.

Integrations

vim/nvim

Press Enter in visual mode to open the first URL in the selection:

" ~/.vimrc
vnoremap <silent> <CR> :<C-U>'<,'>w !squeeze -1 --url --open<CR><CR>

Load every TODO and FIXME of a project into the quickfix list:

:cexpr system("squeeze --todo --fixme --with-location 'src/**/*'")

tmux

Press Enter in copy mode to open the first URL in the selection:

# ~/.tmux.conf
bind -T copy-mode-vi enter send -X copy-pipe-and-cancel "squeeze -1 --url --open"

fzf

Pick a URL from the scrollback of the current tmux pane and open it:

tmux capture-pane -pJ -S - | squeeze --url --uniq | fzf --tac | squeeze --url --open

Shell

List the URLs from your shell history, most recent first:

# ~/.bashrc, ~/.zshrc
urls() { fc -rl 1 | squeeze --url --uniq; }

Performance

No regex engine is involved: SIMD searches and byte classification narrow the positions where a match can start. Big files, streams, and directory trees can be scanned across available cores; -j 1 selects a single thread.

On a generated 56 MiB mixed corpus, single-threaded squeeze used 18 ms of CPU to extract URLs versus ripgrep's 35 ms. The tools use different URL grammars and return different match counts. See the benchmarks for the commands, data, results, and limitations.

Use as a Rust library

Add squeeze-lib, imported as squeeze:

[dependencies]
squeeze = { package = "squeeze-lib", version = "1.0" }

use squeeze::{email::Email, scanner::Scanner, uri::URI};

let scanner = Scanner::new(vec![Box::new(URI::default()), Box::new(Email::default())]);

let line = "Ping ops@example.com about https://example.com/status";
for m in scanner.scan_line(line) {
    let kind = scanner.finders()[m.finder_index].id();
    println!("{kind}\t{}", &line[m.range]);
}

See the API documentation for individual finders and scanner options.

Development

Install mise, then run the same checks as CI:

mise trust
mise install
mise run check

See development and releases for the other tasks and the release process. Bug reports with a sample input that squeeze gets wrong are especially welcome.

License

MIT