Not every service or tool has a first-class integration — but your app can still connect to any Python library directly. By leveraging the built-in Python runtime, the AI Builder can install packages, import libraries, and call their functions from workflows or components. This gives you maximum flexibility to extend your app with the broader Python ecosystem.
When you ask for a certain functionality the AI Builder will first check if there is a first-class integration available. If not, it will search the web to try and find a relevant Python library to fulfill your request. If it finds one, it will install the package and ask you to set any API keys that are required.
There is no built-in integration for Slack. But if you ask the AI Builder to send a message to a Slack channel, it will research itself the best implementation and then use the slack_sdk Python package to send the message.
There is no built-in integration for Scikit-learn. But if you ask the AI Builder to classify some text using scikit-learn, it will research itself the best implementation and then use the scikit-learn Python package to load a pre-trained model and classify the text.
If you are working with a specialized / less well-known library, you can add custom knowledge to help the AI Builder understand how to use it. Simply provide a brief description of the library, its purpose, and example usage in the Knowledge section of your app settings. This will guide the AI Builder when it attempts to call functions from that library.
There are a very small number of libraries that are not supported due to their size. For example, large machine learning frameworks like torch or tensorflow are not supported directly. In these cases, we recommend using a first-class integration that can emulate similar functionality (e.g., the Replicate integration for running ML models in the cloud).