Screenshots
The tool captures an image of a window, a region or the whole display and sends it to a vision model, which reads it the way you would. It works on anything you can see, including images, charts and apps that expose no text.
Guide · Checked 17 October 2026
AI That Can See Your Screen on Mac: 11 Tools Compared
Plenty of Mac apps can now look at your screen and answer questions about it. They differ less in what they can do than in how they look: one capture when you ask, or everything in the background, kept. This guide compares 11 of them on that, and on four other questions that matter more than any feature list.
Disclosure
We make Crodo, which is one of the 11. Every tool here, Crodo included, answers the same five questions from its own documentation, and every entry lists where it falls short. No tool paid to be included. Where something else is the better choice, we say so.
Is there an AI that can see your screen?
Yes. ChatGPT, Claude, Gemini, Raycast and Apple's own Siri AI capture a single window or screenshot only when you ask. A few, like Screenpipe and Highlight, watch in the background and keep a searchable record. That difference matters more than any feature.
At a glance
All 11 compared.
How each one sees your screen, whether it runs in the background, and what happens to what it captures.
| Tool | How it sees your screen | Runs in background? | What happens to the capture | Voice | Price |
|---|---|---|---|---|---|
| ChatGPT | Frontmost window plus its text, on both Command keys | No; opt-in history on Pro, Business and Enterprise | Kept locally in the chat session file | Plus and up | Free plan; Plus $20/mo |
| Claude | A window or region you pick | No | Attached to the chat | Dictation | Free; Pro $20/mo |
| Gemini | Visible part of the frontmost window | No | Saved to Gemini Activity if it is on | Yes | Free |
| Siri AI | A window or region you select | No | Not stored, per Apple | Yes | Free, macOS 27 |
| Raycast | Window text plus a screenshot | No | Not stored | Dictation | Pro from $10/mo |
| VoiceOS | The area you point at | No | Not retained after processing | Yes | From $11.99/mo |
| Crodo | Display under your cursor, on keypress | No | Not stored | Yes | Free |
| Jarvis | A screenshot when summoned | No | Not stored by Jarvis | Yes | Free 40/wk; Pro $16/mo |
| Screenpipe | Event-triggered screenshots plus text | Yes, always | Stored locally, 14 days by default | Dictation only | Free on 1 device; from $25/mo |
| Highlight | Permitted app windows | Yes, by design | Uploaded, then deleted | Yes | From $35/mo |
| Shadow | The meeting window, when it changes | During meetings | Stored locally | Plus | Free; Plus $8/mo |
Checked 17 October 2026from each vendor's own documentation. Plans change often; confirm on the vendor's site before you pay.
How it works
How AI sees your screen.
“Seeing your screen” covers four quite different things. Most tools use one or two of them.
Accessibility text
The tool asks macOS for the actual text inside an app, through the same Accessibility system screen readers use. It is more accurate than reading pixels for long pages and code, and can include text scrolled out of view, but it only works as well as each app exposes its text.
Continuous capture
The tool keeps taking captures in the background and indexes them, so you can search or ask about the past. It is the only approach that answers “what was I looking at on Tuesday”, and the only one that builds up a record of everything.
Agents that act
A few tools go further: they look at the screen, then click, type and navigate in apps you approve. Seeing the screen is how they find their way around.
The decision
On demand or always on: the choice that matters most.
On demand
One capture per question, and nothing in between. Nothing accumulates, so the only exposure is whatever is on screen at the moment you ask. Right for help with what is in front of you: an error, a page, a message, a chart.
Always on
A searchable archive of everything that crosses your screen, including passwords that flash past, private messages and client data. That archive is both the value and the risk. Right when you genuinely need to recall what you saw, and best kept on your own Mac.
The quick test
Do you want help with this screen, or memory of past screens? For help, pick an on-demand tool. For memory, pick an always-on recorder and check where it stores the record.
The list
The 11 tools, grouped by how they look.
Not ranked one to eleven: the right tool depends on whether you want help, memory or meetings. Within each group, the tools most people already have come first.
Tools that answer when you ask
These take one capture at the moment you trigger them and nothing in between. Nothing builds up over time. What differs is how much they see, whether you can talk to them, and where the capture goes.
1
ChatGPT
The biggest name, and the most ways to look
- How it sees your screen
- Press both Command keys and ChatGPT takes an “appshot”: a screenshot of the frontmost window plus that window’s text, read through macOS Accessibility, including text scrolled out of view. In desktop voice mode, saying “take a look at this” does the same. Separately, Computer Use can operate apps you approve inside its Work and Codex modes.
- What happens to the capture
- Appshots are stored locally in the chat’s session file and sent to OpenAI for processing. An opt-in Computer History feature on Pro, Business and Enterprise builds a background record of your activity from accessibility events, not screenshots.
- Voice
- Yes, in the new desktop app, on Plus and higher plans.
- Price
- Free plan available, though OpenAI does not say whether it includes appshots. Plus is $20 a month and is needed for desktop voice. Appshots are not offered on Enterprise or Education plans.
- Needs
- macOS 14 or later, Apple silicon or Intel.
- Try it
- In the new ChatGPT desktop app, open the window you want help with and press both Command keys together.
Best for
A deliberate “look at this window” gesture, and an agent that can actually operate Mac apps.
Where it falls short
One window at a time, never the whole display. Live screen sharing during a voice call exists only in the iPhone and Android apps, not on Mac. Two desktop apps, ChatGPT and ChatGPT Classic, currently coexist with different features.
2
Claude
Show it a region or a window, on any plan
- How it sees your screen
- Quick entry lets you drag over a region or click a window to attach a screenshot to your message, on every plan including Free. Computer use, in beta on Pro and Max, works through screenshots and can operate native apps you approve one at a time.
- What happens to the capture
- Processed in Anthropic’s cloud. Quick-entry captures are attached to the chat like any upload; Anthropic documents no separate retention for them.
- Voice
- Caps Lock dictation through quick entry on macOS 14 or later, plus a voice mode in beta.
- Price
- Free plan includes quick-entry screenshots. Pro is $20 a month, or $17 a month billed annually, and adds computer use.
- Needs
- macOS 13 or later, Apple silicon or Intel.
- Try it
- Open quick entry from the Claude desktop app, then drag across the part of the screen you want to ask about.
Best for
Choosing exactly what to show, down to a dragged region, without paying.
Where it falls short
No continuous awareness: every capture is one you take. Computer use is in beta on Pro and Max only, and Anthropic warns it may struggle with complex multi-step work.
3
Google Gemini
Free, native, and also on both Command keys
- How it sees your screen
- Press both Command keys and Gemini uses what is visible in the frontmost window as context, or pick a window with Share window. Speak to Window (hold fn) captures the active window along with your voice and writes the result into the app you are in.
- What happens to the capture
- If Keep Activity is on, what you share, including screens, is saved to your Gemini Activity, auto-deleted after 18 months by default, and may be used to improve Google’s models, including training. Audio from Speak to Window is not stored.
- Voice
- Yes. Speak to Window works system-wide, in English.
- Price
- Free. Paid Google AI plans raise usage limits, and the local-folder Tasks agent needs AI Pro ($19.99 a month) or Ultra.
- Needs
- macOS 15 Sequoia or later, Apple silicon.
- Try it
- Open the window you want help with, then press both Command keys.
Best for
Fast, free questions about the window in front of you, especially if you already live in Google Workspace.
Where it falls short
Only the visible part of one window. Every Speak to Window use captures the active window, even when you only meant to dictate. Speak to Window is English-only.
4
Siri AI (Apple Intelligence)
Built in, nothing to install, still in beta
- How it sees your screen
- Select a window (Command-Shift-Space) or a region (Command-Shift-6) and ask Siri about it, or Control-click a selection and choose Ask Siri. In apps that support App Intents, Siri also receives structured data about what is on screen, so it can act on it, such as adding an event to Calendar.
- What happens to the capture
- Apple says requests are handled on device where possible and otherwise in Private Cloud Compute, where data is not stored or made accessible to Apple. Training is opt-in.
- Voice
- Siri takes voice, but Apple documents the screenshot flow as typed into the Ask Siri field.
- Price
- Free with macOS 27. Server-side features have daily limits that Apple does not publish.
- Needs
- macOS 27 and a Mac with M1 or later.
- Try it
- On macOS 27, press Command-Shift-Space over a window, then type your question into Ask Siri.
Best for
Anyone on macOS 27 who wants screen help with no extra app and no Screen Recording permission to grant.
Where it falls short
In beta, behind a waitlist, and English-only at launch. Onscreen awareness was announced in 2024 and only shipped with macOS 27 in September 2026.
5
Raycast AI
Real text, not just a screenshot
- How it sees your screen
- Double-tap the right Command key (the suggested hotkey) and Raycast sends the focused window’s text, read through Accessibility, together with a screenshot, so the model works from real text rather than guessing from pixels. You can also attach a selected area or the entire screen.
- What happens to the capture
- Raycast says nothing is captured in the background and nothing is stored: the capture is attached to your message. Your data is not used to train AI models.
- Voice
- Dictation into the chat box, on Pro.
- Price
- Needs Raycast Pro, from $10 a month ($8 a month billed annually). The free plan has no AI.
- Needs
- Raycast v2, which requires macOS 26 Tahoe and Apple silicon.
- Try it
- In Raycast v2, double-tap the right Command key over the window you want to ask about.
Best for
Raycast users who want one hotkey that hands the AI the actual text of what they are reading.
Where it falls short
Pro-only and metered in AI credits. During a chat, the model can take a capture on its own unless you switch that off in settings.
6
VoiceOS
Point at it while you talk
- How it sees your screen
- Hold the Agent Mode trigger and ask out loud while pointing with the cursor or drawing a trail around what you mean, so “this” refers to exactly what you marked. One question can cover up to three monitors. Screen awareness stays off until you turn it on.
- What happens to the capture
- Processed in the cloud by third-party AI providers. Its privacy policy says it does not retain screen content after processing; zero data retention is offered on Enterprise.
- Voice
- Yes, voice-first, with dictation and voice editing too.
- Price
- No free plan; the 7-day trial needs a card. Pro is $11.99 a month billed annually, or $29.99 month to month.
- Needs
- macOS 10.15 or later, per the vendor. Windows too.
- Try it
- Turn on screen awareness in VoiceOS, grant Screen Recording, then hold the Agent Mode trigger and point.
Best for
Pointing at one specific thing on a busy screen without having to explain where it is.
Where it falls short
No free plan. Captures are still images taken when you ask, with no history to look back on. Fully cloud-based, with no on-device option.
7
Crodo
That's usSpoken question and screen on one key, free
- How it sees your screen
- Hold the Read-screen hotkey (Option + Command by default) and ask out loud. Crodo captures the display your cursor is on at the moment you press the key, with its own windows filtered out, downscales it, and sends it with your spoken question.
- What happens to the capture
- The screenshot goes to OpenAI’s API to answer the question. Crodo keeps no copy: it is held in memory and never written to disk or saved to your account. Only the text of your question and the answer is kept.
- Voice
- Yes. You speak the question; the answer appears as text and copies itself to your clipboard.
- Price
- Free, with no usage cap.
- Needs
- macOS 14.2 or later, Apple silicon.
- Try it
- Hold Option + Command, ask your question, and release.
Best for
Asking about anything on screen by voice, for free, in the same app you use for dictation, meeting notes and your inbox.
Where it falls short
English only. Cloud processing, with no offline mode. It answers but cannot click or type for you, and there is no way to point at one part of the screen the way VoiceOS allows.
8
Jarvis
A summon-only bar with connectors
- How it sees your screen
- Summon the floating bar and Jarvis takes a screenshot of what is on screen when you ask. It can then act through more than 30 connectors such as Gmail, Slack, Notion and Linear, with each action set to always allow, ask first, or blocked.
- What happens to the capture
- Jarvis says it does not store screenshots or conversation history; captures are sent to AI providers under those providers’ own terms. Memory is stored locally and encrypted.
- Voice
- Yes: a spoken conversation and dictation.
- Price
- Free for 40 requests a week. Pro is $16 a month ($144 a year); Unlimited is $32 a month.
- Needs
- macOS 14 or later. Windows too.
- Try it
- Summon the bar, grant Screen Recording when asked, and ask about the window in front of you.
Best for
Trying a summon-only screen assistant with connectors at no cost.
Where it falls short
In beta according to its own terms, built by a single founder, with no security certifications yet. Refunds are generally not given.
Tools that watch in the background
These are designed to capture continuously, so they can answer questions about things you saw hours or days ago. That memory is the point, and it is also the risk: everything that crosses your screen ends up in the record.
9
Screenpipe
A searchable local record of everything you saw
- How it sees your screen
- Once installed, it captures in the background, triggered by events such as app switches, clicks, typing pauses and visual changes, pairing each screenshot with the window’s accessibility text. New installs also record meeting audio by default. Other AI tools, such as Claude and Cursor, can query the record through MCP.
- What happens to the capture
- Everything is stored locally. New desktop installs keep media for 14 days by default; plan on around 30 GB a month. Some network features, such as analytics and crash reporting, are on by default.
- Voice
- Dictation in its chat only; it is not a voice assistant.
- Price
- Free on one device for personal, non-commercial use. Paid plans start at $25 a month ($21 a month billed annually).
- Needs
- macOS 13 or later, Apple silicon or Intel; macOS 14.4 or later for system audio.
- Try it
- Install it, grant Screen Recording and Accessibility, and let it run; then ask about your history from its chat, or from Claude through MCP.
Best for
Asking “what was I working on” days later, with the record kept on your own Mac.
Where it falls short
No longer open source: its core moved to a source-available licence in June 2026. Always-on capture records everything on screen, including other people’s messages and meeting audio.
10
Highlight
Watches in the background to draft your follow-ups
- How it sees your screen
- Built to observe your screen in the background: it takes screenshots of permitted app windows and reads their text through Accessibility, then combines that with meeting transcripts and connected apps to draft follow-ups for your approval. Capture can be set to Most Useful, Work Only or Off, with rules by app, site or window, and paused with Incognito Mode.
- What happens to the capture
- Under its new privacy policy, effective 30 October 2026 for existing users, screenshots are uploaded for summarisation and deleted shortly after. AI-written descriptions of your screen activity may be kept, with no stated retention period.
- Voice
- A voice-mode hotkey and audio notes.
- Price
- No free plan. From $35 a month (Lite) to $200 (Max); the 14-day trial needs a card.
- Needs
- Apple silicon or Intel Mac. Windows too.
- Try it
- Start the 14-day trial, grant Screen Recording and Accessibility, and choose a capture level.
Best for
People with back-to-back meetings who want follow-ups drafted from what they actually worked on.
Where it falls short
Now a meetings and follow-up product; the original Highlight app is retired on 15 November 2026. Its own documents disagree about how continuous its capture is, and it does not say how often screenshots are taken.
Tools that capture during meetings
Screen capture tied to a specific trigger: a call. Useful when what matters is what was shown in a meeting, not what is on your screen right now.
11
Shadow
Captures what was shown in your meetings
- How it sees your screen
- During a meeting it automatically screenshots the detected meeting window, only when the content changes, so slides and shared screens end up in your notes. Outside meetings, screen-reading Action Skills run on a keyboard shortcut.
- What happens to the capture
- Meetings, transcripts and screenshots are stored locally as Markdown and images, and transcription runs on your Mac. Screenshots leave the Mac only when an AI or sharing feature uses them.
- Voice
- Yes: Voice Typing, Quick Reply and other Action Skills, on Plus.
- Price
- Free, with unlimited recording, transcription and meeting screenshots. Plus, for screen shortcuts and AI notes, is $8 a month billed annually ($12 monthly).
- Needs
- macOS 14.2 or later, Apple silicon.
- Try it
- Install it, take a call, and check the meeting folder afterwards for the captured slides.
Best for
Bot-free meeting notes that include what was on screen, with everything kept as local files.
Where it falls short
Screen shortcuts are unavailable while a meeting is being recorded, and meeting screenshots stop if the meeting window is hidden. All AI features run in the cloud.
Other options
- ScreenHelp. One keystroke sends a screenshot plus a saved prompt to OpenAI and streams the answer into an overlay. No voice, 30 free answers a month, and an early-access desktop app with no public Pro price.
- Crade. A floating Mac and Windows assistant that snapshots your screen when you ask and can run on your own Claude or ChatGPT account. Free tier, paid from $7.99 a month.
- Clicky (HeyClicky). A Mac-only voice assistant that looks at your screen only when you press its hotkey. Free tier, paid plans at $20 and $100 a month.
- Cluely. A Mac and Windows meeting assistant that listens to calls continuously and reads your screen when you ask. Free tier, Pro at $19.99 a month.
What we left out, and why
- Granola. A meeting notetaker, not a screen assistant.
- Dia. Sees your browser tabs, not the rest of your Mac.
- Rewind. Meta acquired Limitless, the company formerly called Rewind, on 5 December 2025, and Rewind’s screen capture was switched off on 19 December 2025.
- Windows tools. Microsoft’s Copilot Vision and Recall cover similar ground on Windows. This guide sticks to the Mac.
Before you grant anything
macOS permissions, explained plainly.
Every screen tool asks for some of these in System Settings > Privacy & Security. Here is what each one actually lets an app do, and you can revoke any of them there at any time.
| Permission | What it lets an app do | Who asks for it |
|---|---|---|
| Screen & System Audio Recording | Lets an app capture images of your windows or your whole display. | ChatGPT, Claude, VoiceOS, Jarvis, Crodo, Screenpipe, Highlight and Shadow. Optional in Raycast, which can send text only. |
| Accessibility | Lets an app read the text inside other apps, and click or type in them. Raycast notes it is renamed Device Control and Data Access on macOS 27. | ChatGPT, Claude, Raycast, VoiceOS, Jarvis, Crodo (for its hotkeys and pasting), Screenpipe, Highlight and Shadow. Optional in Gemini. |
| Input Monitoring | Lets an app see what you type across all apps. | Screenpipe, optionally. |
| Microphone | Lets an app hear spoken questions and dictation. | Any tool you talk to. |
The exception is Apple's own Siri AI: it is built into macOS 27, and Apple documents no Screen Recording or Accessibility grant for it at all.
How to choose
Which one for which job.
Start from what you are trying to do. One requirement usually settles it.
Explain an error, a setting or a chart
Any on-demand tool works. Free options: Gemini, Claude and Crodo.
Hand over the actual text of a long page or code
Raycast or ChatGPT appshots, which read the app’s text through Accessibility. ChatGPT includes text scrolled out of view.
Point at one thing on a crowded screen
VoiceOS, or drag a region in Claude.
Ask out loud instead of typing
Crodo, VoiceOS, Gemini’s Speak to Window, or ChatGPT voice on Plus.
Have it actually do the task
ChatGPT Computer Use, or Claude computer use on Pro and Max.
Find something you saw days ago
Screenpipe, which keeps the record on your Mac, or Highlight.
Keep what was shown in a meeting
Shadow, free.
Install nothing at all
Siri AI on macOS 27, in beta.
Worth knowing
What these tools can't see.
The gaps nobody puts on a pricing page, and the ones most likely to give you a wrong answer.
One window, or everything
ChatGPT, Gemini and Raycast default to the frontmost window. Claude and Siri AI let you pick a window or a region. Crodo captures the whole display your cursor is on. VoiceOS can span up to three monitors in one question.
Pixels, or real text
Screenshot-only tools read the image. Raycast and ChatGPT appshots also read the app’s text through Accessibility, which is more accurate, but apps that expose little text give them only a screenshot. OpenAI notes that for Google Docs, Gmail, Sheets and Slides, ChatGPT may receive only the visible screenshot.
Gaps in an always-on record
Screenpipe captures on events, with idle snapshots 30 seconds apart on power and up to 120 seconds apart on battery, so a screen that changes quietly can go uncaptured.
At work
Using them on a work Mac.
- Admin controls exist and vary. ChatGPT organisations can turn off voice Screen context, Computer History needs an admin to grant it on Business and Enterprise, and appshots are not offered on Enterprise and Education plans.
- Gemini’s Mac app is on by default for Google Workspace organisations that have Gemini enabled, and is controlled from the Admin console. Its Tasks agent does not work with work or school accounts.
- Apple’s Siri AI can be restricted through Screen Time > Content & Privacy.
- Always-on recorders capture client data and colleagues’ voices along with your own. Screenpipe’s documentation flags the consent questions this raises for other meeting participants.
- The practical rule: check with IT before installing anything that holds Screen Recording permission on a company Mac.
Method
How we checked.
- 01
Every claim comes from the vendor’s own documentation, pricing pages, help centres and release notes, checked on 17 October 2026.
- 02
A second pass then tried to disprove each claim against those same primary sources, and corrected it wherever it succeeded.
- 03
Every tool answers the same five questions: how it sees the screen, whether it runs in the background, what it keeps, where captures go, and what it costs.
- 04
Prices are US list prices. Local prices vary, and these plans change often, so confirm on each vendor’s site before you pay.
- 05
This is a documentation-based comparison, not a timed hands-on test.
- 06
We make Crodo. It answers the same five questions as everything else, with its limitations listed. No tool paid to be included.
Questions
The ones people ask most.
Is there an AI that can see my screen on a Mac?
Yes. ChatGPT, Claude, Google Gemini, Raycast, VoiceOS, Jarvis and Crodo can all look at your screen when you ask, and Apple’s Siri AI does it natively on macOS 27. Screenpipe and Highlight go further and capture your screen in the background. The biggest difference between them is not features but whether they look only when you ask or watch all the time.
Does an AI that sees my screen record it all the time?
Most do not. ChatGPT, Claude, Gemini, Siri AI, Raycast, VoiceOS, Jarvis and Crodo take a single capture when you trigger them and nothing in between. Screenpipe and Highlight are designed to capture continuously in the background, and Shadow captures automatically during meetings. ChatGPT also offers an opt-in Computer History feature on paid plans, which records accessibility events rather than screenshots.
Can these tools see my passwords?
They can see anything visible in what they capture. On-demand tools only see a password if it is on screen at the moment you ask, so close sensitive windows first. Always-on recorders can capture whatever passes across the screen, which is why Screenpipe and Highlight offer per-app exclusions and pause controls, and why Screenpipe’s own documentation warns that redaction can miss things.
How do I let ChatGPT see my screen on a Mac?
In the new ChatGPT desktop app, open the window you want help with and press both Command keys together; ChatGPT attaches a screenshot of that window plus its text. In voice mode, turn on Screen context in settings and say “take a look at this”. In ChatGPT Classic, use the + button and choose Take Screenshot. Live screen sharing during a voice call is only available in the iPhone and Android apps.
What is the best free AI that can see your screen?
Google Gemini’s Mac app, Claude’s quick entry and Crodo are all free and can see your screen when you ask. Apple’s Siri AI is free with macOS 27 but is in beta behind a waitlist. Jarvis gives 40 free requests a week. Raycast’s screen feature needs a paid Pro plan, and VoiceOS and Highlight have no free plan.
Do any of these work without sending my screen to the cloud?
Not the on-demand assistants: ChatGPT, Claude, Gemini, Raycast, VoiceOS, Jarvis and Crodo all send captures to cloud models. Apple handles some requests on device and sends larger ones to its Private Cloud Compute. Screenpipe stores its record on your Mac, though some of its network features are on by default, and Shadow keeps meeting captures as local files.
Can an AI see my screen and then do the task for me?
Yes, in two cases here. ChatGPT’s Computer Use and Claude’s computer use (in beta, on Pro and Max) can see the screen and click and type in apps you approve, asking permission app by app. The rest, including Crodo, answer questions about what they see but do not operate your apps.
Do I need to give Screen Recording permission?
For most third-party tools, yes: it is what lets an app capture images of your windows. Raycast can work with Accessibility alone, sending text without a screenshot. Apple’s Siri AI is built into macOS and needs no Screen Recording grant at all. You can revoke any app’s access in System Settings > Privacy & Security.
Does screen-watching AI slow down my Mac?
On-demand tools only do work when you trigger them. Always-on recorders run constantly: Screenpipe lists 600 MB to 4 GB of memory depending on settings, and around 30 GB of storage a month.
What happened to Rewind AI?
Meta acquired Limitless, the company formerly called Rewind, on 5 December 2025, and Rewind’s screen and audio capture was switched off on 19 December 2025. Screenpipe and Highlight are the closest background-capture alternatives on Mac.
Go deeper