add-image-vision

To process WhatsApp image attachments for Claude multimodal input, the system downloads the images, resizes them using sharp, encodes them to base64, and delivers them as multimodal content blocks alongside text for agent interpretation.

NanoClaw agents interpret WhatsApp group chat images by receiving the downloaded, resized, and base64-encoded image attachments as multimodal content blocks passed directly to Claude for analysis and response generation.

Enabling image vision for NanoClaw agents requires the sharp dependency for image resizing, alongside modifications to source files like src/image.ts and integration across agent-runner and container-runner code.

Yes, base64 encoded images can be sent to Claude through a containerized agent runner by integrating image processing logic into the container-runner code to deliver multimodal content blocks.

NanoClaw agents fail to respond to WhatsApp images without image vision capabilities enabled, as they require specific image processing logic to download, resize, and encode attachments into multimodal blocks for Claude.

The best way to add image understanding to a WhatsApp agent workflow is implementing an image processing pipeline that downloads attachments, resizes them with sharp, and encodes them as base64 multimodal content for Claude.