Show your screen to AI.

AI sees the screen, decides, and acts.
Screen capture built for that. One CLI.

# Show the whole screen to AI
$ gaze capture
 
# Zoom into the region that matters
$ gaze capture --area 340,120,800,500
 
# Send text instead of an image
$ gaze ocr --area 340,120,800,500
 
# Record a clip of the interaction
$ gaze record --duration 10

From Claude Code, Cursor, or a shell script.

Screenshot
Show as an image
Full screen, a selected area, or a single window.
OCR
Show as text
Convert the screen to text. 1,600 image tokens become 80.
Record
Show as a video
Record a flow. Frames are extracted automatically.
CLI
AI looks directly
From Claude Code or Cursor, AI captures the screen itself.

For moments like these.

Show a UI to AI and have it fixed

Capture the full screen, let AI locate the part that matters, zoom in by coordinates, and iterate with higher precision. No more pasting screenshot after screenshot.

As image
1,600 tok
As text
80 tok
Tokens 95% less
Processing fully local
Languages Japanese · English

Pass an error screen as-is

Capture the error screen from the CLI. Run OCR and the stack trace comes through accurately.

Capture a flow in one go

Burst mode takes 5 or 10 frames from a single command, ready to hand to AI together.

Mask sensitive data before sharing

Auto-detect and mask emails, API keys, and personal data. Internal screens stay safe to hand to AI.

Show motion with a recording

Record the screen and extract frames. For conveying animations or transitions.

A workflow that runs without hands

Crop the region, run OCR, save it. Layout detection and coordinate targeting are handled by AI.

Watch the screen on a schedule

gaze watch observes the screen and, on change, pulls the diff via OCR. Good for deploy checks and UI drift.

AI sees, decides, and acts.

Capture the full screen, locate the right region, zoom by coordinates, read it as text, move on.
The "see and decide" loop humans used to run — now AI runs it through the CLI.

# AI checks the full screen
$ gaze capture
→ screenshot saved: /tmp/gaze-001.png
 
# Zoom into the region by coordinates
$ gaze capture --area 340,120,800,500
→ screenshot saved: /tmp/gaze-002.png
 
# Read it accurately as text
$ gaze ocr --area 340,120,800,500
→ "Error: Cannot read property 'id' of undefined
    at UserService.getUser (src/user.ts:42)"
 
# Mask sensitive data, then convert to text
$ gaze capture | gaze mask | gaze ocr

Local processing. Nothing leaves the device.

All on-device

Capture, OCR, resize — all local.

Sensitive data removed automatically

Mask emails and API keys before anything goes to an LLM. Admins can block sending outright.

Security review stays short

No image data leaves the device, so the DPA scope is narrow. Most checklist items come back "not applicable".

Plans

Free
$ 0
50 captures a month. OCR, CLI, and MCP included.
  • Area and window capture
  • OCR text extraction
  • Copy to clipboard
  • Global shortcut
Start for free
Pro — subscription
$ 39 /year
From $4.99/mo. Always on the latest version.
  • Everything in Pro
  • Auto-updates to the latest version
  • Cancel anytime
  • From $4.99/mo
Start subscription

One-time license updates after year one: $15/year. Skip renewal and your current version keeps working.

Team and Enterprise

Admins configure masking rules and outbound controls.

Team

$12 /user/month

From 3 seats. Annual: $120/user/year.

  • Everything in Pro
  • Team admin console
  • Shared configuration
  • Usage analytics
  • Priority support
  • Volume discounts
Start Team plan

Enterprise

Contact us

SSO, audit logs, dedicated support. For larger rollouts.

  • Everything in Team
  • SSO (SAML/OIDC)
  • SCIM provisioning
  • Audit logs
  • Data retention policy
  • On-prem LLM support
  • Dedicated CSM
  • SLA
Contact sales

Frequently asked

How is this different from the built-in macOS screenshot?

The built-in tool just saves an image. Gaze adds OCR, burst capture, and sensitive-data masking, callable from the CLI in one command.

Where does the data go?

Nowhere. Processing is local. Telemetry is off by default.

How accurate is OCR?

It uses the macOS Vision framework. Japanese is supported.

What happens when the one-time update window ends?

Your last version keeps working. Renewals are optional at $15/year.

Can I use both CLI and GUI?

Yes. brew install gaze gives you the CLI; launching the app gives you the GUI. Same binary.

Does it work offline?

Yes. OCR, capture, and masking all run locally. No network required.

Try it.

brew install gaze. 50 captures a month, free.