Vision

The underlying model has no native image understanding. When an image arrives, do not use the Read tool — run scripts/vision.js to get a text description back.

Requires Node.js 18+ (native fetch) and network access to an OpenAI-compatible vision API.

When to use

• User shares an image path (local file or remote URL)
• Message contains "Saved attachments:" listing image files
• User asks to analyze, describe, or identify image content

Quick start