base.json

{
  "model": "gpt-4o-mini",
  "response_format": {
    "type": "json_schema",
    "json_schema": {
      "strict": true,
      "name": "svg_response",
      "schema": {
        "type": "object",
        "properties": {
          "input_description": {
            "type": "string",
            "description": "Description of input, including interpretation of what is being asked and interesting features"
          },
          "input_features": {
            "type": "array",
            "description": "List of features in the input, including their description and coordinates",
            "items": {
              "type": "object",
              "properties": {
                "description": {
                  "type": "string",
                  "description": "Description of feature"
                },
                "top_left_x": {
                  "type": "number",
                  "description": "Top-left corner X coordinate of feature"
                },
                "top_left_y": {
                  "type": "number",
                  "description": "Top-left corner Y coordinate of feature"
                },
                "bottom_right_x": {
                  "type": "number",
                  "description": "Bottom-right corner X coordinate of feature"
                },
                "bottom_right_y": {
                  "type": "number",
                  "description": "Bottom-right corner Y coordinate of feature"
                }
              },
              "required": ["description", "top_left_x", "top_left_y", "bottom_right_x", "bottom_right_y"],
              "additionalProperties": false
            }
          },
          "output_description": {
            "type": "string",
            "description": "Description of response, both in general and specifics about how the response can be represented in an SVG overlayed on the screen. Include specifics such as position in coordinates of response objects"
          },
          "svg": {
            "type": "string",
            "description": "An SVG in correct SVG format which will be drawn on top of the existing screen elements"
          }
        },
        "required": [
          "input_description",
          "input_features",
          "output_description",
          "svg"
        ],
        "additionalProperties": false
      }
    }
  },
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "You are a helpful assistant. You live inside of a remarkable2 notepad, which has a 1404x1872 sized screen which can only display black and white. Your input is the current content of the screen. Look at this content, interpret it, and respond to the content. The content will contain both handwritten notes and diagrams. Respond in the form of a JSON document which will explain the input, the output, and provide an actual svg, which we will draw onto the same screen, on top of the existing content. Try to place the output in an integrated position. Use the `Noto Sans` font-family when you are showing text.\nDo not use a style tag tag. Do not use any fill colors or gradients or transparency or shadows. Do include the xmlns in the main svg tag."
        }
      ]
    }
  ],
  "max_tokens": 3000
}