Skip to content

look_at Tool

Description

The look_at tool extracts basic information from media files (PDFs, images, diagrams) when a quick summary suffices over precise reading. It's designed for simple text-based content extraction without requiring the full Read tool.

⚠️ Important Limitations: - NEVER use for visual precision, aesthetic evaluation, or exact accuracy - Use the Read tool instead for cases requiring visual precision or exact accuracy - This tool is read-only and does not modify any files

Usage

look_at(
    file_path: str,  # Absolute path to the file to analyze
    goal: str        # What specific information to extract from the file
)

# OR for clipboard/pasted images:
look_at(
    image_data: str, # Base64 encoded image data
    goal: str        # What specific information to extract from the image
)

Parameters

Parameter Type Description Required
file_path string Absolute path to the file to analyze Yes (if no image_data)
image_data string Base64 encoded image data (for clipboard/pasted images) Yes (if no file_path)
goal string What specific information to extract from the file Yes

Examples

Extract text from a PDF

look_at(
    file_path="/path/to/document.pdf",
    goal="Extract the main headings and key points from this document"
)

Analyze a screenshot

look_at(
    file_path="/path/to/screenshot.png",
    goal="Identify the main UI elements visible in this screenshot"
)

Analyze a clipboard image

look_at(
    image_data="base64_encoded_data_here",
    goal="Extract any visible text from this image"
)

Technical Details

  • Implementation: Uses multimodal model analysis via the registered multimodal-looker model
  • Model: Uses the agent's configured vision-capable model (typically google/antigravity-gemini-3-flash)
  • Output: Returns clean text content extracted from the file
  • Performance: Optimized for quick analysis rather than precise extraction

When to Use

✅ Good for: - Quick text extraction from PDFs - Basic content analysis of images - Simple diagram understanding - Fast content summaries

❌ Not for: - Visual precision tasks - Aesthetic evaluation - Exact layout analysis - Complex document processing - Any task requiring pixel-perfect accuracy

Best Practices

  1. Be specific in your goal parameter to get the most relevant information
  2. Use Read tool when you need precise, detailed analysis
  3. Combine with other tools for comprehensive analysis workflows
  4. Handle errors gracefully - the tool may fail on complex or unsupported file formats

Error Handling

The tool may return errors for: - Unsupported file formats - Corrupted files - Permission issues - Model resolution failures - Network connectivity issues (for cloud-based analysis)

Always check the error output and fall back to the Read tool if look_at fails to provide adequate results.