Skip to content
Prompt Presets

Free for macOS Predictions + prompt insertion for coding chats

Predict your next prompt. Insert it in one shortcut.

Prompt Presets suggests the three prompts you will want next, and drops those, or any prompt from your library, into your coding chat. Works with T3 Code, Codex, Claude Desktop, Claude Code, Cursor, Devin and GitHub Copilot.

Watch how it works

Free, local-first. No account required. ⌃⌥⌘P⌃⌥⌘T Ctrl+Alt+PCtrl+Alt+Shift+T

Features

A colorful workspace for the prompts you actually reuse.

See each part running, in the real app.

Predictions

Three next prompts, one keypress away.

Press ⌃⌥⌘P in your chat app. A small popup appears beside your mouse with three suggested next prompts, each tagged with complexity, effort, cost and risk so you know what you are asking for before you ask. Click one, or press 1–3, to insert it.

  • Pick your engine: your Codex CLI (ChatGPT subscription), your Claude Code CLI (Claude subscription) or an OpenAI API key. With Pro, the hosted Fast Prompt Presets engine needs no setup.
  • Context is bounded: by default up to 10 recent prompts and the last 3 answers from one session, capped at 16,000 characters.
  • File paths, session IDs and chat titles are never sent to the model.

The chips are the engine’s estimates, not measurements.

Predictions — Prompt Presets
Screen recording: pressing the prediction shortcut over a T3 Code chat opens a popup with three suggested next prompts, each with complexity, effort, cost and risk chips; pressing 2 inserts the second one into the chat draft.

Prompt insertion

Your best prompts, a keystroke away.

Press ⌃⌥⌘T for a popup of your top prompts, ranked by how often you use them. Press 1–9 to insert. A prompt with {{variables}} turns into a small form first, so the one part that changes is the only part you type.

  • It appends to the draft already in the box and never presses Send.
  • Give any prompt its own global shortcut.
  • No safe place to insert? It copies the prompt instead and tells you so.
Top prompts — Prompt Presets
Screen recording: the top-prompts popup lists saved prompts ranked by use; clicking one appends it to the chat draft without sending.

Retrieval

Your best prompts are already in your history.

Prompt Presets automatically collects what you have already typed in your local T3, Codex, Claude Code, GitHub Copilot, Cursor and Devin histories into a prompt list. Keep the good ones in your library, dismiss the rest.

  • Injected system and context envelopes are filtered out, so you only see what you wrote.
  • Histories are opened read-only (SQLite read-only mode).
  • Retrieved prompts land in library.json on your machine. Only prompts you keep can ever be synced, and only with Pro.
Prompt list — Prompt Presets
Screen recording: prompts retrieved from local T3 and Codex histories appear in the prompt list; one is opened and added to My Prompts.

Teams · Pro

Share a prompt. Keep your own copy.

Share a saved prompt snapshot to a team. Teammates save a linked copy, and when the author shares again they see “Update available” and choose explicitly: keep their edits or replace them.

  • Sharing is explicit: nothing is shared until you press Share saved snapshot.
  • Local edits are never overwritten without an explicit Keep or Replace.
  • Saved copies keep working offline.
Teams — Prompt Presets
Screen recording: sharing a saved prompt to a team, and a teammate seeing an Update available badge on their linked copy.

How it works

Three steps. No setup wizard.

  1. Step 1: Install

    Download the app and open it. Your library is a single file on your computer.

  2. Step 2: Grant Accessibility (optional)

    It lets Prompt Presets see which chat is active and append to its prompt field. macOS asks only when you click “Enable access…” in Settings.

  3. Step 3: Press the shortcut

    In your chat app, press ⌃⌥⌘P for predictions or ⌃⌥⌘T for your top prompts. Pick one with a click or a number key.

Integrations

Meets your chats where they live.

What works where depends on what each app exposes on your machine. Anything that cannot be verified falls back to copying.

  • T3 Code Retrieve · Predict · Insert
  • Codex Retrieve · Predict · Insert
  • Claude Desktop Predict · Insert
  • Claude Code Retrieve · Predict
  • Cursor Retrieve · Predict · Insert
  • Devin Retrieve · Predict · Insert
  • GitHub Copilot Retrieve · Predict · Insert
  • Any app via Other exports

Predictions work from any of these chats: the shortcut detects the chat on screen, or you pick it in Prompt Presets. In Cursor that is the open tab of the Agents window, in Devin the open local session. Insert appends to the draft only where the app’s composer can be verified, and otherwise copies the prompt. Claude Desktop means local Code sessions; cloud chats in Claude, Cursor and Devin are not read.

Privacy

Your prompts are yours.

Built so there is as little to trust as possible. The details are in the privacy policy.

  • Local-first

    Your library is one file, library.json, on your machine. Prompts leave your computer only when you ask for a prediction or use an optional Pro feature.

  • Read-only histories

    Chat histories are opened with SQLite in read-only mode. Prompt Presets never writes to another app’s data.

  • Keys in the OS keychain

    An OpenAI API key, if you use one, lives in your system’s credential store (macOS Keychain, Windows Credential Manager or the Linux Secret Service). Never in the library, exports or histories.

  • No telemetry unless you opt in

    Anonymous usage stats are off until you switch them on, and they never include prompts, chats, paths or credentials.

  • A locked-down webview

    The desktop app’s webview runs under a strict CSP and makes no network calls itself. Every request goes through Rust.

  • You pull the trigger

    Predictions run only when you ask for them. They are cancellable, time-limited and never retried on their own.

Pricing

Free to use. Pro to go faster, together.

Everything that runs on your computer is free. Pro adds fast hosted predictions and the parts that need a service.

Free

€0 forever

Everything local.

  • Your prompt library, with variables and global shortcuts
  • Top prompts popup and insertion into your chat apps
  • Predictions with your own Codex CLI, Claude Code CLI or OpenAI key
  • Retrieval from your local chat histories
  • Browse the curated prompt catalog and save copies

Yearly

€19 / year

Everything in Free, plus:

  • Fast predictions: 1,500 a month, three suggestions in under a second, no key to set up
  • Teams: share prompt snapshots and keep linked copies
  • Account sign-in with Google, Apple, GitHub or email
  • Sync of your prompts across computers
See Pro

Founder Lifetime

Best value

€99 one-time

Everything in Yearly, with lifetime access, plus:

  • 50,000 fast predictions in total, no monthly reset
  • Every future Pro feature, no upgrade fees
See Founder Lifetime

All allowances reset on the 1st of each month. Founder predictions count once generation starts, even when interrupted. After the allowance, keep predicting with your own API key or CLI engines.

FAQ

Questions, answered straight.

Something missing? Write to [email protected].

Does it press Send for me?

Never. Inserting a prompt appends it to whatever is already in the composer and stops there. If Prompt Presets cannot verify a safe place to insert, it copies the prompt and says so, leaving your draft untouched.

What does it send to the model?

Only when you ask for predictions: by default up to 10 recent prompts and the last 3 answers from the one chat you are in, capped at 16,000 characters, sent to the engine you chose (your Codex CLI, your Claude Code CLI, the OpenAI API or, with Pro, our Fast Prompt Presets service). File paths, session IDs and chat titles are not included. Retrieval and insertion send nothing anywhere.

Does it need the Accessibility permission?

It is optional, and it is what lets Prompt Presets detect the active chat and append to the prompt field. Without it, saved prompts still open as drafts in supported apps or get copied. macOS asks only when you click “Enable access…” in Settings.

Which platforms are supported?

macOS 13 or later (Apple Silicon and Intel) is the platform to rely on today. Builds for Windows and Linux exist but are still in validation. Inserting directly into another app’s prompt field depends on what each desktop allows: on Windows it is implemented but not yet tested on real machines, and on GNOME and KDE under Wayland Prompt Presets copies the prompt instead.

Where is my data?

In library.json, written atomically with owner-only permissions: ~/Library/Application Support/Prompt Presets/ on macOS, %APPDATA%\Prompt Presets on Windows, ~/.local/share/Prompt Presets on Linux. Your prediction key, if you use one, is in your system’s credential store (macOS Keychain on a Mac). An account and sync are optional Pro features, and only exist if you turn them on.

Do I need an account?

No. The library, shortcuts, predictions with your own engine, retrieval all work without one. Accounts (Google, Apple, GitHub or email) come with Pro, for Teams and sync.

How do I cancel Pro?

Yearly is a subscription sold through Lemon Squeezy, so you cancel it with Lemon Squeezy; its receipt emails normally link to where you manage it. Founder is paid once, so there is nothing to cancel. In the app, Deactivate frees the license on that computer.

Stop retyping the same prompt.

Free, local-first, and out of your way until you press a key.

Other platforms