Computer API Preview

Since its popularization by the Apple Macintosh in 1984, the graphical user interface has become the dominant medium for human-computer interaction. Multimodal language models can understand these interfaces, but struggle to "point" to visual controls with high precision.

The Open Interpreter Project has developed (and here freely hosts) an API which is capable of locating visual controls with single-pixel precision.

To get the coordinates of any on-screen control, provide its natural language description via the "query" parameter, along with a representation of the computer's display via the "base64" parameter.


Python Example

import requests

query = 'blue bell icon'
base64 = 'data:image/png;base64,...'

response = requests.post(
  'https://api.openinterpreter.com/v0/point/',
  params = {'query': query, 'base64': base64}
)

response.json() # Outputs [(x, y)], a list of relative coordinates to each instance of the icon


This API was designed in Seattle for code interpreting language models. Last modified January 1st, 2024