add-image-vision

To process WhatsApp images for Claude multimodal ingestion, the system uses sharp to resize the downloaded image attachments and converts them into base64 encoding. This prepares the images as efficient multimodal content blocks that the agent can reliably interpret.

Yes, you can send WhatsApp screenshots directly to a Claude agent by applying image handling logic that downloads the attachments and converts them into Claude-ready multimodal inputs. The agent can then interpret the screenshot content and respond appropriately in-line during the conversation.

Processing WhatsApp images for multimodal models requires base64 encoding to transform the raw image data into a text-based format that can be passed reliably within multimodal content blocks. The sharp library handles the image resizing before this encoding step occurs.

Before adding image vision to a WhatsApp agent, you need to ensure your container and agent-runner support base64 multimodal content blocks. You must also install the sharp dependency, apply the documented merge steps, and rebuild the container before restarting the service.

WhatsApp image attachments may not work if the container or agent-runner lacks support for base64 multimodal content blocks. You must sync agent-runner source caches, rebuild the container, and restart the service after applying the image handling modifications to ensure proper routing.