AI Agents
Last updated
Was this helpful?
Last updated
Was this helpful?
CARTO AI Agents enhance user interactions with Builder maps by making data exploration intuitive and conversational. With AI Agents, users can extract insights and interact with maps seamlessly through a conversational interface.
Admin users can enable CARTO AI Agents for their organization by navigating to the AI Agents tab under Customizations in the organization settings. Once activated, Editor users can add an AI Agent to their Builder maps, which will be accessible in Viewer mode.
Once AI Agents are enabled at the organization level, Editors can activate them on individual Builder maps. To do this, navigate to the AI tab in the left panel of the Editor mode in Builder and toggle the option to enable the AI Agent. Once the Agent is created, you can interact with it as an Editor using the Preview mode or as a Viewer when you share your map.
Please note AI Agent is not supported in Public maps.
When the AI Agent is created or updated, the Agent is automatically provided with the map's configuration details—such as the sources, layer configuration, widgets, and tools available for end-users. This ensures the Agent can effectively understand and interact with the map without requiring manual input of these details.
Customizing the Agent is optional but highly recommended to enhance its functionality and align it with specific use cases. Editors can customize the Agent by adding:
Map context: Providing additional instructions and context specific to the map. For example, include business logic, style preferences, or any relevant information to improve the Agent’s responses.
Conversation starters: Predefining up to four prompts to guide user interactions and help users get started.
User guide: Including a brief introduction that explains the Agent’s functionality and how users can interact with it.
When accessing your map, the AI Agent icon is located at the bottom of the map interface. Clicking on it opens the conversational dialog, where users are greeted with a brief description and any predefined conversation starters.
During this interaction the Agent may temporarily control the map to perform tasks such as filtering, zooming, or switching layers.
CARTO AI Agents can provide both text-based insights and direct map interactions. Available tools include:
Search and zoom to locations: Easily identify and zoom to locations, including specific geographic extents like "Northern Portugal" or "Berlin city center."
Extract widget values: Analyze data conversationally by extracting widget values.
Filter data via widgets: Apply conversational filters to widgets, simplifying data exploration.
Switch layers on and off: Toggle layers directly from the conversation.
Retrieve maps coordinates: Access the latitude and longitude of your current map position.
Limitations
Extracting widget values is capped at 20 values for Category widgets and 50 values for Histogram and Time Series widgets.
Table Widget data extraction is not supported.
Filtering via Time Series widget or SQL parameters is not yet available.
Each organization has a monthly quota for CARTO AI Agent usage, which resets on the first day of each month. The quota includes:
Threads: Each organization can initiate up to 250 threads per month.
Tokens per thread: Each thread is limited to 250,000 tokens (words, punctuation, and formatting).
CARTO AI Agents leverage the OpenAI APIs with the GPT-4-o model. Our plan is to integrate AI Agents with additional AI providers and models, further expanding their functionality and versatility.
Other future advancements will include:
Allow AI agent to query sources for richer insights, enabling real-time data retrieval and transforming raw geospatial data into actional insights without relying just on widgets.
By leveraging CARTO Workflows, users will be able to define analytical and data processing pipelines, empowering AI Agents to deliver richer and more meaningful results.
Extending AI Agent capabilities to include Public maps, broadening their applicability and accessibility.
By enabling this feature, you confirm acceptance of the .
For more information about token limits, refer to .
