The Best macOS Launcher? I Compared Look, Raycast, Alfred and Spotlight (and Look Won)

Full disclosure before we start: I built one of the four apps in this post. This is the kind of article where the referee is also playing in the match.

So I am going to be genuinely fair to the other three. If I inflate their weaknesses, you have no reason to believe anything I say about mine either.

Why launchers deserve this much thought

Cmd+Space is probably the key you press most on a Mac. I have never counted. I suspect counting would be upsetting.

Everything goes through that door: opening an app, finding a file, doing quick math, jumping to a settings pane. If that door is half a second slow, you lose about a minute a day. A minute sounds like nothing. The real cost is the small "ugh" you feel while waiting. That part adds up in a way minutes do not.

The short answer

Look.

Fast. Free. Open source. Fully local. AI that runs on your own machine. Same app on macOS, Windows and Linux.

No other launcher gives you all six. Plenty give you one or two. Nobody gives you all of them. That is why I built it.

That said, the others genuinely win in some situations:

• You need a deep extension store to wire up your company's tools → Raycast
• You have ten years of hand made Workflows that are now real assets → Alfred
• You truly do not want to install anything → Spotlight

Everything outside those three cases, I think Look is the right pick. Here is why.

The comparison table

	Look	Raycast	Alfred	Spotlight
Price	Free, all of it	Free + paid plan	Free + Powerpack (one off)	Built in
License	Open source (GPLv3)	Closed	Closed	Closed
Platforms	macOS / Windows / Linux	macOS / Windows (beta)	macOS only	macOS only
Account	Not required	Needed for some features	Not required	Apple ID
Stays local	Yes	Partly (cloud sync)	Yes	Partly (web results mixed in)
Telemetry	None	Yes	None	Sent to Apple
Background daemon	None	Yes	Yes	System resident
AI	On device (Apple Intelligence / Ollama)	Cloud LLM (paid)	None	Siri / ChatGPT

Six rows won outright, two tied. I made the table, so grade the objectivity accordingly.

Spotlight: it is already there and it costs nothing

That is a real advantage. Zero setup, zero decisions. It wins on that alone for a lot of people.

But use it daily and the friction accumulates.

You type "bluetooth" because you want the Bluetooth settings pane. That is the whole goal. What comes back on top is some random document with the word bluetooth in it, or a web search suggestion. The thing you actually wanted is two arrow keys down.

Worse, the order changes from day to day for the same query. So you end up looking at the screen every single time to confirm. If you have to look at the screen, it is not really a keyboard launcher anymore.

There is also the part where you wanted a local file and got news, a dictionary entry and Wikipedia. You can turn some of that off in settings. Not all of it.

This is the app that pushed me into writing my own, so, you know, consider the source.

Raycast: I am not going to pretend I can beat the ecosystem

The extension store is enormous. Linear, Notion, GitHub, Jira, internal tools. Whatever you use, something probably exists for it. The UI is polished and it feels good from the moment you install it. Credit where it is due.

The tradeoffs:

• Extensions run on a Node.js runtime, so memory usage is permanently non trivial
• Cloud sync and AI features require an account, and the best parts sit behind the paid plan
• It is closed source, so you cannot verify what it does with what you type

If your job is "connect my work tools and go faster on a company Mac", Raycast is still the first thing I would recommend. I recommend it to people regularly.

Alfred: still standing after more than a decade

That fact alone says a lot. It is local, it does not chatter over the network, and it is rock solid. Buy the Powerpack and you unlock Workflows, where shell scripts and AppleScript stack up into automation nobody else has. If you are deep into that, you do not need another launcher.

The weak spots: macOS only, and the UI feels a bit classical by current standards. Also the paywall sits at the entrance rather than at the advanced end, so the free version stays close to a plain app launcher.

Look: this is the part where I stop being modest

Look is a keyboard first, local first launcher. Native SwiftUI on macOS, search engine written in Rust, Tauri on Windows and Linux. No Electron anywhere.

1. Speed was the constraint, not the goal

Under 1 ms per query on a 2000 item index. 30 microseconds to browse with an empty query. Holding those numbers is exactly why the app is a Rust core under a native shell. The flip side is that plenty of other things got sacrificed to keep them. More on that near the end. There is no background daemon.

2. Free, and open source under GPLv3

There is no paid tier. There is no "that feature is Pro". Partly on principle, partly because I do not want to run a subscription business.

The source is all on GitHub, so what it reads and what it sends is verifiable rather than promised. No account, no telemetry.

3. The AI runs on your machine

Apple Intelligence, or whatever model you have pulled in Ollama.

• Natural language file search ("pdfs from last week")
• Summarize, translate or rewrite whatever you copied
• Calendar and reminders in plain language ("move my dentist to friday")
• Type > and the panel becomes a conversation. Every change previews first, Enter confirms, Cmd+Z undoes

The preview step and Cmd+Z were there from day one. Watching a model confidently create the wrong calendar event once is all the design research you need on that subject.

Personal context (your calendar, your clipboard, your saved facts) is held to a stricter rule: it is attached only when inference is running on this machine. Point Look at a remote model and none of it goes out unless you explicitly allow it in settings.

4. The tools are already inside

No extensions to install. Cmd+/ and they are there.

• calc: factorials, constants, and notation like 1920x1080 handled correctly
• pomo: pomodoro timer with a menu bar readout and a background music folder
• todo: tasks grouped by day, completion rates, and a GitHub style year heatmap
• speed: internet speed test on a live dial, which I built that way because it looks cool
• kill: process killer that also finds things by port (:3000)
• Clipboard history (c"), translation (t"), dictionary (tw")

5. Super actions

With an empty query you get a control strip: Wi-Fi, Bluetooth, mic, theme, keep awake, screensaver, weather, battery, now playing, restart, shut down.

Cmd+W for Wi-Fi, Cmd+B for Bluetooth. It visibly cuts how often you open System Settings. Restart and Shut Down need two presses to fire. There was a version where one press was enough, and I learned why that was a bad idea through direct personal experience.

6. Add your own rows

Drop a TOML file in ~/.look/sources/ and the contents of a directory, the lines of a file, or the output of a command get indexed and ranked next to your apps and files. Git repos, SSH hosts, docker containers, project folders.

No plugin store, no JavaScript. Ready made definitions live in lookbook: copy one in, reload, done.

7. Same app on your other machines

Windows and Linux run the same core (Alt+Space). Mac at work, Linux at home, and your muscle memory transfers intact.

Where Look is weak

Leaving this out would make the rest less believable.

• The extension ecosystem does not come close to Raycast's. There is no store. There is no plan to build one
• Natural language AI actions need an Ollama setup (the deterministic paths, like the @ syntax, work with no model at all)
• Development moves fast, so behavior does shift between versions
• It is one person. File an issue when you hit a bug and it will probably be me reading it

Install

brew tap kunkka19xx/tap
brew install --cask look

Then go to System Settings > Keyboard > Keyboard Shortcuts > Spotlight, unbind Cmd+Space there, and give it to Look. Skip this step and you will conclude the app does not launch. (I have made this mistake. On software I wrote.)

Release builds are signed and notarized, so no Gatekeeper workarounds.

Verdict

Raycast wins on extensions. Alfred wins if you already own a decade of Workflows. I will happily concede both.

Everything else, Look wins.

Sub millisecond search. No paid tier. Full source. Zero telemetry. No account. No daemon. AI on your own machine. The same app across three operating systems. All of it free.

I never accepted that paying monthly for a moderately fast launcher was supposed to be the normal state of things. So I built the alternative. Try it once and Cmd+Space stops being something you can give back.

GitHub: https://github.com/kunkka19xx/look
Docs: https://noah-code.com/docs/look

A star on the repo makes the next release arrive sooner. That is not a figure of speech.