Vision (describe_image)

The describe_image tool lets the agent analyze images you attach or reference by path.

Requirements

• Code or Office mode (when vision feature is enabled)
• VISION_API_KEY or provider config supporting image models (see config.toml)
• Image within size limits enforced by the runtime

Typical uses

• Screenshot UI bugs → describe layout and suggest fixes
• Diagram or whiteboard photo → extract structure
• Scan a chart in inbox/ for an Office summary

How to invoke

• Drag/drop or paste an image into chat, or
• Place an image under workspace and ask the agent to read it

The model receives a text description from the vision backend, then continues reasoning with other tools.

Privacy

Images are sent to your configured vision provider — treat sensitive screenshots accordingly. Do not attach credentials or personal ID photos unless you accept that risk.

Related: File tools · API key