Inventory is where you create and manage the objects that monitor you AI deployments. There are three objects: Profiles, Use Cases, and Models.

A model is the connection to an underlying large language model. A profile is the policy applied to traffic, controlling which signals are detected and what happens to inputs and outputs. A use case represents a deployed AI application and ties a profile to it.

Where you start depends on what you want to do.

• If you want to monitor live traffic, you need all three. The model provides the LLM connection, the profile governs the traffic, and the use case represents the application being monitored.
• If you want to test individual prompts, go to Playground. Playground runs in listen mode without a model, so you can send a prompt and see the signals detected on the input. Connect a model when you also want the model's response back along with the signals detected on that response.
• If you want to test a collection of prompts in bulk, use Cannon. A Cannon run processes a collection of prompts through a profile, so you need a profile and a collection to run one.

These paths share one rule. A model is what produces a response from the LLM. Without a model, traffic is still analyzed on the input side, which is what list mode does. With a model connected, a profile can be set to enforce, and you also get the response and its analysis. you select a model within a profile, and a model is required wherever you need a response back: a profile in enforce mode, Playground, Cannon, and live API traffic.

A model has to exist before a profile can connect to one to enable enforce mode, so creating a model comes first when your goal involves a response from the LLM. Beyond that, the objects do not have to be built in a fixed order.

Inventory contains a separate list page for each object type: Models, Profiles, and Use Cases. Each page displays existing objects and the option to create a new one.