Fuzzy search helper for Alfred 3+ workflows

Author: Dean Jackson

Last Updated: 12/12/19 21:01:02

Stars: 42

Get Workflow Get Source Code View on GitHub

Fuzzy search for Alfred

fuzzy.py is a helper script for Alfred 3+ Script Filters that replaces the “Alfred filters results” option with fuzzy search (Alfred uses “word starts with”).

How it works

Instead of calling your script directly, you call it via fuzzy.py, which caches your script’s output for the duration of the user session (as long as the user is using your workflow), and filters the items emitted by your script against the user’s query using a fuzzy algorithm.

The query is compared to each item’s match field if it’s present, and against the item’s title field if not.

Example usage

fuzzy.py only works in Script Filters, and you should run it as a bash/zsh script (i.e. with Language = /bin/bash or Language = /bin/zsh).

Instead of running your own script directly, place ./fuzzy.py in front of it.

For example, if your Script Filter script looks like this:

/usr/bin/python myscript.py

You would replace it with:

# Export user query to `query` environment variable, so `fuzzy.py` can read it
export query="$1"
# Or if you're using "with input as {query}"
# export query="{query}"

# call your original script via `fuzzy.py`
./fuzzy.py /usr/bin/python myscript.py

Note: Don’t forget to turn off “Alfred filters results”!


Grab the Fuzzy-Demo.alfredworkflow file from this repo to try out the search and view an example implementation.


Fuzzy search, and this implementation in particular, are by no means the “search algorithm to end all algorithms”.


By dint of being written in Python and using a more complex algorithm, fuzzy.py can only comfortably handle a small fraction of the number of results that Alfred’s native search can. On my 2012 MBA, it becomes noticeably, but not annoyingly, sluggish at about ~2500 items.

If the script is well-received, I’ll reimplement it in a compiled language. My Go library for Alfred workflows uses the same algorithm, and can comfortably handle 20K+ items.


Fuzzy search is awesome for some datasets, but fairly sucks for others. It can work very, very well when you only want to search one field, such as name/title or filename/filepath, but it tends to provide sub-optimal results when searching across multiple fields, especially keywords/tags.

In such cases, you’ll usually get better results from a word-based search.

Technical details

The fuzzy algorithm is taken from this gist by @menzenski, which is based on Forrest Smith’s reverse engineering of Sublime Text’s algorithm.

The only addition is smarter handling of non-ASCII. If the user’s query contains only ASCII, the search is diacritic-insensitive. If the query contains non-ASCII, the search considers diacritics.


You can tweak the algorithm by altering the bonuses and penalties applied, or changing the characters treated as separators.

Export different values for the following environment variables before calling fuzzy.py to configure the fuzzy algorithm:

Variable Default Description
adj_bonus 5 Bonus for adjacent matches
camel_bonus 10 Bonus if match is uppercase
sep_bonus 10 Bonus if after a separator
unmatched_penalty -1 Penalty for each unmatched character
lead_penalty -3 Penalty for each character before first match
max_lead_penalty -9 Maximum total lead_penalty
separators _-.([/ Characters to consider separators (for the purposes of assigning sep_bonus)

Multiple Script Filters

If you’re using multiple Script Filters chained together that use different datasets, you’ll need to set the session_var environment variable to ensure each one uses a separate cache:

# Script Filter 1
export query="$1"
./fuzzy /usr/bin/python myscript.py

# Script Filter 2 (downstream of 1)
export query="$1"
export session_var="fuzzy_filter2"
./fuzzy /usr/bin/python myotherscript.py


The fuzzy matching code was (mostly) written by @menzenski and the algorithm was designed by @forrestthewoods.