Cartobot integrates with the OpenAI API to allow GPT-based models to assist a user with their query,
returning a result as JSON that is programmatically parsed to execute the user’s request.
Commands to interact with the Cartobot application take the form of natural language (e.g. “go to
Paris,
please”).
A user enters text in the “Type a Message” textbox and hits Enter or presses Send. Users can also
click
Start Recording,
enter a voice command, and then click Stop Recording to send the voice command to the AI. When a
user
records a command,
the audio is sent to OpenAI where a model called Whisper parses the audio as text, and then sends
that
to a GPT-based model.
Navigation:
Map navigation can be bucketed into a few methods: Flying to a location, panning in a direction, and zooming in and out. When navigating, a marker will be placed on the map at the location.Sample prompts:
“take me to Tokyo”
“go to the statue of liberty”
“fly to the largest city by population in the United States”
Styling:
Currently, the application only supports styling a single layer at a time and the layer name must be in the prompt. Use the dropdown at the top of the page to see a list of available layers.Sample prompts:
“change the building color to light brown”
“turn the water to a light blue color”
“make the forests a pastel color reminiscent of Autumn”
“make the water harder to see”
“change opacity of parks to 45%”
To use interpolated functions, such as "Make water dark blue when at zoom 10 and light blue at zoom 17" The zoom level must be in ascending order, or else the function will not operate.
Label Styling:
There is some basic label manipulation functionality in Cartobot as well which includes changing label sizes and color, as well as halo size and blur amount. If the map includes icons, the size of icons can also be adjusted.Sample prompts:
“change the city labels to size 45”
“change the halo size of country labels to 20”
"increase size of place labels"
"make place label halo color dark green"