Phantom Review (2026): Pricing, Features & Honest Verdict

Phantom Review: The Verdict

Phantom is a voice-first AI agent for macOS that lives in your Mac’s notch and operates the machine for you: clicking, typing, launching apps, moving files and firing shortcuts on command. The control is real. The capacity limits are undefined. Best for: Mac power users who want hands-free control of their machine. Price: free plan, paid from $20/mo. Rating: 7/10.

We found Phantom at #16 on Product Hunt with 110 upvotes, which undersells what it does. Most “AI for your Mac” launches turn out to be dictation with a nicer icon. This one takes an outcome and goes and performs it.

What is Phantom?

Phantom is a macOS app built by Benaja Heger that parks an AI agent behind your Mac’s notch. You speak or type from anywhere in the system, describe the result you want, and Phantom plans the steps and carries them out across whatever apps are involved. It reads the screen and app context relevant to your request, decides a sequence of actions, executes them, then verifies the outcome. Nothing opens. No window steals focus. The company line is that it “turns thought into action with no interface in between”, and for once that is a description of the mechanic rather than marketing.

What separates this from the wave of Mac AI tools that launched over the past year is the verb. Wispr Flow and Superwhisper turn speech into text and hand the text to you. Raycast gives you a command palette with a model attached. Phantom takes actions on your behalf: it types, clicks, scrolls, launches apps, invokes Shortcuts, works with files, and pushes actions out to services you have connected. Under the hood it runs on OpenAI and ElevenLabs, which is worth knowing before we get to the privacy claims. Phantom launched on Product Hunt in July 2026 and categorised itself under AI voice agents and productivity.

What Are Phantom’s Key Features?

Notch-resident, zero-interruption input

The interface is the notch, and that decision does more work than it sounds like. Every competing product costs you a context switch: a window appears, your current app loses focus, you type, you read, you dismiss it, you find your place again. Phantom collapses that into speaking at your laptop while your hands stay where they were. For anyone who lives in one app for hours at a time and needs to fire off a task without leaving it, this is the difference between a tool you use and a tool you forget you installed.

Outcome-level execution across apps

You give Phantom the result, not the recipe. It handles planning, works across application boundaries, and then verifies that what it did matches what you asked for. That last step is the interesting one. Plenty of computer-use agents will happily click seven things, miss the eighth, and report success. Building verification into the loop is the correct design, though it is also the part hardest to judge from a marketing page, and the part most likely to be inconsistent in practice on a product this young.

Action Policy permissions

Phantom ships an in-app Action Policy that governs what the agent is allowed to do without asking. You set the permissions, and actions outside them need your approval. This matters more than a normal permissions screen, because the blast radius here is your entire machine. An agent with click and type access to a logged-in Mac can send an email, empty a folder, or authorise something in a browser tab you forgot was open. Making the policy user-editable rather than a single all-or-nothing toggle is the right call, and it is the feature we would want to inspect most closely before granting broad access.

On-demand context instead of continuous capture

Phantom does not sit there screenshotting your screen on a timer. It pulls context when a request needs it. That is a meaningful architectural difference from the always-recording category of Mac AI tools, and it is the strongest privacy claim on the page. Be clear about what it does and does not mean, though: on-demand still means your screen contents leave the machine when a request needs them, and they go to OpenAI and ElevenLabs. This is less exposure than continuous capture, not zero exposure.

How Much Does Phantom Cost?

Phantom starts at $0 and paid plans begin at $20/mo. There are three tiers: Free at $0/mo on shared capacity with starter features, Pro at $20/mo with more capacity, and Max at $100/mo with the highest capacity. Annual billing takes 20% off, so Pro works out at roughly $16/mo and Max at about $80/mo, or $192 and $960 a year respectively.

Here is the problem, and it is the single biggest flaw in the product as sold. “Capacity” is never defined. Not in requests, not in minutes, not in actions, not in tokens. Free is “shared capacity”, Pro is “more”, Max is “highest”. You cannot work out whether Pro covers your usage without buying Pro and finding out, and the gap between $20 and $100 is a 5x jump with no stated unit to justify it. Prefactor got criticised in this same slot for not publishing spans per run, and at least Prefactor published the span counts. Phantom publishes no number at all.

On price positioning, Pro at $20/mo sits above the Mac AI tools it will be compared against. Raycast Pro is $8/user/mo billed annually or $10 monthly. Superwhisper Pro is around $8.49/mo, with a $249.99 lifetime licence. Wispr Flow Pro is $15/mo, or $12/mo billed annually, on top of a free tier capped at 2,000 words a week. So Phantom is charging between a third and two and a half times more than the field.

That premium is defensible, because those three tools do not do what Phantom does. Dictation is not agency. Paying $20 for something that completes a multi-app task is a different purchase from paying $8 for faster transcription. But Max at $100/mo needs to justify itself against a Claude or ChatGPT subscription at $20 to $30, and without a published capacity number, nobody can run that comparison.

Who is Phantom Best For?

Use Phantom if you work on a MacBook all day, across many apps, and you keep hitting the same friction: knowing exactly what you want done and resenting the twelve clicks it takes. Designers, developers, operators and anyone who runs their working life through a keyboard will get the most from it. The free tier costs nothing to test against your real workload, which is exactly how you should evaluate this.

Use Phantom if voice is your preferred input and you have found existing dictation tools frustrating because they stop at text. Being able to say the outcome and have the machine do it is a real capability gap that Raycast and Superwhisper do not close.

Skip Phantom if you handle sensitive material. Screen context going out to OpenAI and ElevenLabs is a hard blocker for regulated work, client confidential files, or anything under an NDA that names cloud processors. The on-demand architecture reduces the exposure but does not remove it, and there is no self-hosted or local-model option.

Skip Phantom if you need predictable costs. Undefined capacity tiers are a budgeting problem for a business, and a $100/mo ceiling with no stated unit is not something a finance team will approve. And treat the notch dependency as an open question: the interface premise is the notch on recent MacBooks, and the site does not spell out the experience on a Mac that does not have one, or on an external display.

Best Phantom Alternatives

Raycast is the incumbent for Mac keyboard-driven productivity, at $8/user/mo billed annually or $10 monthly, with a capable free tier and a Teams plan at $12/user/mo. Its extension ecosystem is enormous and its AI features are solid. What it will not do is take autonomous action across your apps: you still drive, it just gets you to the controls faster. Cheaper, more mature, less ambitious.

Wispr Flow is the best-known voice input tool, free on a Basic plan capped at 2,000 words per week, then $15/mo monthly or $12/mo annually, with Teams at $10 to $12 per user. It is excellent at turning speech into clean, context-aware text anywhere on your system. It is not an agent. If what you actually want is to stop typing rather than to stop clicking, Flow is the cheaper and more reliable purchase.

Superwhisper is the privacy pick, with a free tier and Pro at about $8.49/mo, $84.99/yr, or $249.99 once for a lifetime licence. Its selling point is local transcription models, so your audio can stay on your Mac. That is the opposite trade to Phantom: maximum privacy, minimum capability. The lifetime option also makes it the cheapest of the group over a three-year horizon.

Final Verdict: Is Phantom Worth It?

Phantom is doing something the rest of the Mac AI category is not. Voice input into an agent that operates the whole machine, wrapped in a permissions model you control, sitting in the one piece of screen real estate nobody else was using. The design instincts here are good: on-demand context rather than constant recording, verification after execution, an editable Action Policy rather than a single trust toggle. For a first launch from one founder, that is a coherent product with a real point of view.

The reasons to hold back are the pricing page and the maturity. Selling three tiers separated by an undefined word is not acceptable for a paid product, and it will cost Phantom conversions from exactly the careful buyers it should want. Add a solo founder, a dependency on two external AI providers, and a category where reliability is everything and a 90% success rate feels like failure, and this is not yet something to route your working day through. Install the free tier, give it narrow permissions, and run it on tasks you can afford to have gone wrong. If it holds up over a fortnight, $20/mo is fair. Do not pay for Max until somebody at Phantom writes down what capacity means.