Skip to main content

​
Overview

The Read tool supports images and PDFs for Claude’s vision and document understanding capabilities. File type is detected automatically based on extension, and content is base64-encoded for the LLM.

​
Supported Formats

TypeExtensionsMax Size
Images.png, .jpg, .jpeg, .gif, .webp5MB
PDFs.pdf32MB
TextAll othersNo limit (supports offset/limit)

​
Usage

File type is detected by extension — no special handling needed: 
// Images - automatically base64-encoded
handle.send_input("What's in screenshot.png?").await?;

// PDFs - automatically base64-encoded
handle.send_input("Summarize quarterly-report.pdf").await?;

// Multiple files
handle.send_input("Compare design-v1.png and design-v2.png").await?;

​
Tool Result Types

The framework uses different content types internally: 
pub enum ToolResultData {
    Text(String),                    // Regular text files

    Image {                          // PNG, JPEG, GIF, WebP
        data: Vec<u8>,
        media_type: String,          // "image/png", etc.
    },

    Document {                       // PDF files
        data: Vec<u8>,
        media_type: String,          // "application/pdf"
        description: String,
    },
}

​
Prompt Caching

Images and PDFs support prompt caching automatically: 
// First request: Image cached
handle.send_input("Analyze screenshot.png").await?;

// Second request within 5 min: Cache hit (90% savings)
handle.send_input("What colors are used in screenshot.png?").await?;
The last content block gets cache control applied automatically. This reduces costs when analyzing the same files multiple times.

​
File Attachments

Images and PDFs can also be sent via attachment tags: 
let input = r#"Analyze this error:
<vibe-work-attachment>./error.png</vibe-work-attachment>"#;

handle.send_input(input).await?;
See File Attachments for details.

​
Limitations

  • Extension-based detection: File type is determined by extension, not file contents.
  • No streaming: Files are read completely before processing.
  • No image generation: The SDK only supports reading/analyzing images.
  • Provider support: Image/PDF analysis requires Claude (AnthropicProvider). Other providers may not support these features.

​
Next Steps

File Attachments

Attachment tags for files

Prompt Caching

Cache images/PDFs for cost savings