Phantom

Phantom

Phantom · Writing

Phantom is a voice- and text-driven AI assistant for macOS. You give it an outcome, and it plans the steps, works across your apps, and checks the result. It leans on screen context only when a task needs it, and every action runs under the permissions and Action Policy you set. No other assistant on the Mac quite works this way.

Interface preview of Phantom

About Phantom

What Is Phantom

Phantom is an AI assistant built for macOS and nothing else. You talk to it or type, and it carries out work inside the apps already on your Mac. Tell it to send a message, fill a form, or move files, and it figures out the steps instead of making you click through them.

The point is hands-free execution, not another chat window. Phantom can type, click, scroll, launch apps, run shortcuts, touch files, and send actions to services you connect. A task like "email the report to my manager" gets planned and executed across the tools involved, then verified. That's the whole pitch.

The catch is trust and scope. Because Phantom acts on your machine, it asks for macOS permissions and an Action Policy before doing much. It doesn't watch your screen all the time. It does take screenshots and read Accessibility context when a task needs them, and it stays inside the access you grant.

Getting Started

  1. Download Phantom from the official site and install it on a Mac running a current version of macOS.
  2. Sign in with your Google account so your plan and settings carry over.
  3. Grant the macOS permissions Phantom requests, such as Screen Recording and Accessibility, then set an Action Policy level you're comfortable with.
  4. Speak or type a task, review the plan Phantom proposes, and let it run.
  5. Adjust permissions or the Action Policy any time in System Settings or inside Phantom if an upcoming task needs tighter limits.

Product Information

A quick look at Phantom's pricing, supported platforms, and performance.

Free PlanYes
Paid Plans$0 - $100/mo
PlatformmacOS
DeveloperPhantom
CategoryWriting
Release DateApr 2025
Latest UpdatedJul 2025
Website VisitsN/A
Website Global RankN/A
API AvailabilityN/A

Best for

The users, tasks, and scenarios where this tool fits best.

Users

  • Mac power users who juggle many apps and want chores handled without leaving the keyboard.
  • People with limited mobility or repetitive strain who prefer voice and dictation over clicking and typing.
  • Teams and solo workers who already run Google accounts and want a task runner tied to their Mac.

Tasks

  • Sending routine messages and emails
  • Moving and organizing files
  • Filling tedious web or app forms
  • Launching and steering apps

Scenarios

  • Clearing a daily backlog of small actions that eat your morning, like scheduling, replying, and filing.
  • Working hands-busy on the Mac, such as following a recipe or handling hardware while dictating tasks.
  • Early project setup where the same few steps happen across mail, calendar, and documents.

Key features

Voice and Text Control

You start every task by saying or typing what you want, not by hunting through menus. Dictation covers both quick commands and longer spoken notes, so you can draft an email out loud and send it in one flow. For anyone who types slowly or thinks faster than they type, this is the feature that changes how the Mac feels to use. Speak it, and move on.

On-Demand Screen Context

Phantom reads what's on screen when a task calls for it, using screenshots and Accessibility data rather than running a constant recorder. That means it can act on the page or window in front of you without you copying text or describing layout, which saves the fiddly back-and-forth that makes most automations break. It also matters for privacy: the site states it doesn't continuously inventory your apps or windows, and context comes up only when a submitted task needs it.

Apps and Connected Services

Phantom works across macOS apps that expose standard controls, and with supported services you choose to connect. So it can handle a task that starts in mail, continues in a calendar, and finishes in a document without you stitching the steps together, which is exactly the kind of multi-app errand that normally breaks a single-purpose helper. The available actions depend on the app, the service, and the permissions you've granted.

Action Policy and Permissions

You pick how much authority Phantom has. Observe only lets it explain and annotate your screen. Limited local lets it open apps and scroll but blocks clicks, typing, and file changes. Autonomous lets it act without a per-action prompt inside the limits you set, which is the mode most people want once they trust the assistant with their everyday apps and files. The policy doesn't override macOS permissions or connected-service rules. You stay in charge of the blast radius.

Task Planning and Verification

Give Phantom the outcome, not the recipe. It lays out the steps, runs them, and checks whether the result matches what you asked for. This planning step separates it from a plain dictation tool. It can sequence several actions and confirm the job finished. That verification is the part most voice tools skip.

On-Device and Cloud Processing

Phantom splits the work between your Mac and the cloud. Local processing handles what it can, and cloud-backed features kick in only when a task uses them. The trade-off is capacity. Heavier or more frequent tasks draw on your plan's cloud capacity, which is why the tiers differ.

Pros and cons

Pros

  • Runs tasks across real Mac apps instead of only chatting, so it fits existing workflows.
  • The three Action Policy levels let you start cautious and loosen control as you trust it.
  • Screen context arrives on demand and only with permission, which is a lighter approach than always-on recording.
  • A free starter tier means you can test whether the voice-first model suits you before paying, and since tasks run through the same standard controls the paid tiers use, the trial tells you a lot about long-term fit.
  • It works through standard macOS controls, so it doesn't demand special integrations from every app. Not bad for a young product.

Cons

  • macOS only. Windows, Android, and iOS users can't use it at all.
  • Useful cloud capacity sits behind the Pro ($20/mo) and Max ($100/mo) tiers, and starter capacity is shared, so heavy users will likely pay.
  • Because it can click and type on your behalf, mistakes or an over-broad permission set carry real risk. Review irreversible actions yourself.
  • There's no confirmed public API, so you can't wire this AI agent for Mac into your own tools or scripts.

Frequently asked questions

It takes a task you speak or type, plans the steps, and carries them out across your apps by typing, clicking, scrolling, launching apps, running shortcuts, and working with files or connected services. Think of it as a screen context AI that watches the task, not your whole desktop. You control which of those actions it's allowed to take.