A Mac focus stack for deep work: keyboard-first, AI-free

Most "focus app" recommendations are about blocking. Block Twitter. Block Reddit. Block your own browser. The result is a Mac that looks like a parental-controlled iPad — useful, but not the same Mac you bought.

This post takes the opposite approach: instead of blocking what distracts you, configure your Mac so the friction of distraction is structurally higher than the friction of work. Every tool in this stack is local, keyboard-driven, and free of AI features whose surfaces interrupt rather than help. The result is a Mac that is quietly easier to work in for two-hour deep-work blocks.

The stack is opinionated. Most pieces have credible alternatives; I'll name them. The point is the shape of the configuration, not the specific app names.

The principle

Distraction is a friction differential. If "open Twitter" takes 200 ms and "find the file I was editing" takes 4 seconds, your brain takes the easier path. The fix is to make the work-path faster than the distraction-path. Five concrete moves do most of the work:

A launcher with sub-200ms cold open for the work-files you open often.
A keyboard-driven editor and terminal so you don't reach for Dock icons.
A clipboard history so you stop losing the thing you copied.
A focus mode that mutes notifications during work blocks.
A timer / focus session tool that defines the work block.

Each piece is small. Together they tilt the friction landscape.

Layer 1: Launcher

A launcher you can summon in one keystroke replaces the "open Finder, navigate to the folder, double-click the file" sequence. The keystroke is the same for every file; navigating Finder is different for every file. The launcher wins on friction every time.

Picks for 2026:

Lume — local-first, fast cold open, no AI panel that wants your attention.
Raycast — capable; consider disabling its AI surfaces if you want fewer interruptions.
Alfred — quiet by design, well-suited to a focus stack.

The criterion for this stack specifically: the launcher should not have features that ping for your attention. A launcher with a "what's new" badge, a "try the new AI features" hint, or a chat panel that wants you to engage is the wrong tool for deep work. The launcher should be a silent service that opens what you ask for and disappears.

Layer 2: Editor and terminal

The editor and terminal are where the work itself happens. The focus criterion is the same: minimal chrome, keyboard navigation, no panels that nudge you.

Editor: Neovim, Helix, Zed, or VS Code with the chrome stripped (close the sidebar, close the panel, ⌘B and ⌘J as your toggles).
Terminal: Ghostty, Alacritty, or Apple Terminal. All three are fast and quiet.

The configuration discipline: hide everything you do not actively use. The minimap, the inline-suggestion chip, the breadcrumb bar, the welcome page. Each takes a small slice of attention. The cumulative quiet is worth the small one-time setup.

For AI assistance, the focus stack uses local models (Ollama) only. The reason: cloud AI suggestions arrive asynchronously, sometimes seconds after you have moved on, which is the worst possible interruption. Local AI is faster and produces fewer "wait, what's that?" moments. Detail: Local LLM on a Mac vs cloud-powered launcher AI.

Layer 3: Clipboard

A clipboard history is a focus tool, not just a productivity tool. The reason: without history, you have to remember "I need to keep this on the clipboard until I paste it," and that remembering is a cognitive tax. With history, you copy aggressively and retrieve as needed.

Pick: Maccy. Free, local, fast, configurable to 20-200 items, ⌘⇧V to open. The full landscape: macOS clipboard history options.

Layer 4: Focus mode (notifications)

macOS Focus modes are underused. They are also good. The Tahoe 26 Focus app:

Define a "Work" focus.
Allow notifications from: nobody, or one or two specific people (your manager, your partner).
Allow apps: Calendar, the editor. Deny everything else.
Schedule: 09:00–12:00, 14:00–17:00 on weekdays.
Allow critical alerts: yes (these are safety / security alerts you actually need).

Once configured, the Focus mode runs on its own. Slack, Mail, Messages, browser notifications — all silent during work blocks. Your phone (if signed into the same Apple ID and Focus syncing is on) goes quiet too.

This is the single biggest distraction-reduction lever on a Mac. Most users have heard of Focus and ignored it. Configure it once and the daily attention savings are real.

Layer 5: Timer / focus session

A timer is the loop that defines a work block. Without a timer, "I should focus" never starts and never ends. With a timer, the work has a shape: 50 minutes on, 10 minutes off, three times in a morning.

Picks for 2026:

Be Focused — free, simple, in the menu bar.
Flow — free, also in the menu bar.
Session — paid, more polished, includes session journaling.
macOS built-in Stopwatch and Timer (Clock app) — minimal but works.

The picks differ in polish; functionally they all start a count-down and tell you when it ends. Pick one, bind it to a launcher trigger, use it daily.

The deep-work consensus on session length: 45-90 minutes of focused work, 10-15 minutes of break. Pomodoro's 25/5 is fine if you are starting out; longer sessions tend to produce more once you are warmed up.

What you deliberately leave out

The focus stack is opinionated by omission. The things not in it, and why:

Distraction blockers

Freedom, Cold Turkey, Self Control, Opal — all credible products. They work by adding a hard friction to distracting sites: you cannot access X for the duration of the block.

I do not include them in the stack because, for me, they create a cat-and-mouse dynamic. Once you bypass the blocker once, the blocker is dead. If your focus depends on the blocker being absolute, the focus is fragile.

The stack above tries instead to make the work-path more attractive. If you find that insufficient, add a blocker — it is a fine addition, not a replacement for the rest of the stack.

Cloud productivity dashboards

Notion-style dashboards, paid task managers with team features, "command center" apps. The deep-work stack works against you when it requires a five-app sync to know what to do next. A plain Markdown file with today's three tasks is more durable.

AI nag panels

Any tool that interrupts to suggest things — "Try the new feature!", "You have unread suggestions" — is unwelcome in this stack. Disable or remove. This is the single hardest one to enforce because vendors keep adding these surfaces; the discipline is to audit each tool quarterly and silence what you can.

Daily flow with the stack

A morning with this setup, in detail:

08:55 — Mac wakes. Focus mode activates automatically at 09:00. Slack, Mail, browser notifications silent.

09:00 — Open the project. ⌘Space, type three letters of the repo name, ↩. Editor opens to the file I had open last.

09:05 — Start the timer. 50-minute session.

09:05–09:55 — Code. Terminal in one half of the screen, editor in the other. Maccy holds the last 30 copies. Local LLM provides completions. No notifications, no Slack pings, no mail badge.

09:55 — Timer ends. Notification (the one macOS-native one I allow). 10-minute break. Walk away from the desk.

10:05 — Second session. Same shape.

11:00 — Done with the morning's two sessions. Cool down, switch to less-deep work for the rest of the morning.

The shape of the day is consistent because the tools are consistent. No "what should I work on now?" because the timer started a block of pre-decided work. No "is this notification important?" because Focus mode silenced it. No "where did I leave that command?" because the clipboard history has it.

What the stack costs

Money: $0 to ~$50 for paid alternatives. Maccy free, Rectangle free, Lume free, Ollama free, Be Focused free. Optional upgrades: Session ($5), Paste ($30/yr), Raycast Pro ($10/mo if you want their AI).

Time: One afternoon to set up, including the Focus mode configuration. The dotfile-able pieces (editor, terminal, launcher config) survive future migrations.

Attention: The stack stops asking for your attention by design. Once configured, it disappears.

How to know it's working

A few signals that the stack is paying off:

You stop checking Slack at 09:30. Not by willpower; Focus mode silenced it.
The timer's end-of-session ding is sometimes a surprise. Means you were absorbed.
You stop opening tabs you didn't decide to open. Because the friction-shape no longer rewards it.
The afternoon energy crash is smaller. Deep work is less exhausting than shallow context-switching, despite the intuition.

These are subjective signals. They show up after a week or two of consistent use, not on day one.

When to add or subtract

The stack is a starting point. Adjust:

Add a blocker if attention-friction is genuinely insufficient (some apps, some people, some life phases need the harder fence).
Add a journal if you find yourself ending sessions without knowing what you did.
Subtract the timer if you have natural rhythms and the timer feels like overhead.
Subtract the focus mode if your work requires you to be reachable (only you can decide).

The stack is a default; your daily reality is the data.

A small philosophical note

The reason I framed this stack as "AI-free" is not anti-AI. I use AI tools daily — local Ollama for completions, occasional cloud Claude for harder tasks. The "free of AI features whose surfaces interrupt" qualifier is specifically about chat panels, suggestion popovers, and badge nudges that arrive in the middle of a focus session.

Useful AI is the kind you summon. Distracting AI is the kind that summons you. The focus stack chooses tools where AI is the former, not the latter.

That distinction generalizes: any tool whose features pull at you, no matter how clever, is a focus liability. Any tool that quietly does what you ask and disappears is a focus asset. Build the stack from the second category.