glance-sdk 0.1.0

Structured screen understanding for AI apps. Replace screenshots with instant, accurate UI data.

glance-sdk

Structured screen understanding for AI apps. glance.run

Replace expensive screenshot-to-vision-model pipelines with one function call. Glance reads the macOS accessibility tree and gives your AI structured data about every UI element — role, label, value, and exact pixel coordinates.

Install

pip install glance-sdk

Usage

from glance_sdk import screen, capture, find, check_access

# LLM-ready text — drop directly into your prompt
context = screen()

# Full structured data
state = capture()
print(state["app"])           # "VS Code"
print(state["elementCount"])  # 342
print(state["captureTimeMs"]) # 28.5

# Find element by name → exact coordinates
btn = find("Submit")
print(btn["centerX"], btn["centerY"])  # 520, 340

# Check accessibility permission
has_access = check_access()

Why

	Screenshot	Glance
Speed	2-5 seconds	~0.8 seconds
Tokens	~3,000 (image)	~500 (text)
Cost	~$0.03 / call	~$0.001 / call
Positions	AI guesses	Exact from OS

Requirements

• macOS 14.2+
• Accessibility permission (System Settings -> Privacy & Security -> Accessibility)

License

MIT