Help Center

AI sees the screen and clicks any UI element by natural language description. Uses 2-pass refinement for screen-edge accuracy.

Parameters: prompt (required), button (Left/Right/Middle), clickType (Single/Double), provider, model, outputX, outputY

Prompt examples:

• "Click the Save button"
• "Click the File menu in Finder"
• "Click the red close button on the top-left of the window"
• "Click the search field in Safari"

Tip: Be specific about location and appearance. "the blue Submit button in the bottom right" works better than "Submit".

AI monitors the screen at regular intervals and waits until a described condition is met. Replaces fixed-time waits with intelligent visual checks.

Parameters: prompt (required), timeoutMs (default: 30000), pollingMs (default: 2000, min: 500), provider, model

Prompt examples:

• "The download progress dialog has closed"
• "The Notes app window is visible on screen"
• "The loading spinner has disappeared"
• "A confirmation dialog with an OK button is showing"

Tip: Use specific, visually verifiable conditions. "The Notes app window is visible on screen" is better than "Notes is open".

AI reads text from the screen and stores it in a variable. Works with any font, any language, any app.

Parameters: prompt (required), outputVariable (default: AiOcrResult), provider, model

Prompt examples:

• "Read the total price shown on screen"
• "Read all text in the Notes app"
• "Read the error message in the dialog"
• "Read the title of the active window"

AI checks the screen and returns true/false. Ideal for verifying that an action succeeded before proceeding.

Parameters: prompt (required), outputVariable (default: AiValidateResult), provider, model

Prompt examples:

• "The file was saved successfully"
• "The login page is showing"
• "An error message is displayed"
• "The checkbox is checked"

Autonomous agent that takes full control of the screen to complete a described task. Dual-provider Computer Use: Anthropic (Claude Sonnet 4.6 / Opus 4.7) or OpenAI (GPT-5.5 / GPT-5.4 / computer-use-preview). The AI captures screenshots, reasons about the next action, clicks, types, scrolls, and repeats until the task is done.

Parameters: prompt (required), maxSteps (default: 30, max: 200), timeoutSeconds (default: 120, max: 600), outputVariable, provider, model

Prompt examples:

• "Open the Calculator app and compute 2 + 5"
• "Open TextEdit, type Hello World, and save to Desktop"
• "Open System Settings and enable Dark Mode"
• "Open Safari, navigate to apple.com, and take a screenshot"

Note: Requires an Anthropic or OpenAI API key. Choose the default model in AI Settings (Sonnet 4.6 / Opus 4.7 / GPT-5.5 / GPT-5.4 / computer-use-preview). GPT-5.5 is Tier 1 friendly and fast; Sonnet 4.6 is reliable for complex layouts; computer-use-preview requires Tier 3+. Cmd+Space (Spotlight) is unavailable via automation — the AI uses Launchpad to open apps.

Send a text prompt to any LLM and store the response in a variable. Text-only (no screen capture). Works with Apple Intelligence.

Parameters: prompt (required), outputVariable, provider, model

Prompt examples:

• "Summarize the following text in 3 sentences: {=myText}"
• "Translate to English: {=japaneseText}"
• "Generate a professional email reply to: {=emailContent}"
• "Calculate the tax for {=price} at 10%"

A Boolean (hexagonal) block that uses AI Vision to evaluate true/false conditions. Use inside if/while blocks for visual condition branching.

Parameters: prompt (required), provider, model

Prompt examples:

• "The login screen is showing"
• "There is an error message on screen"
• "The file download is complete"