A model defines the connection to the underlying large language model. It is one of the three objects you set up in Inventory: a model is the LLM connection, a profile is the policy applied to traffic, and a use case ties them to a specific application. You select a model within a profile.

A model is required only when you need a response back from the LLM. This applies to a profile in enforce mode, the Playground, the Cannon, and live API traffic. A profile in listen model does not require a model.

Models are created from the Models list page in Inventory.

Create a Model

1. Navigate to Inventory > Models and click Create Model.
2. Enter a Name. This field is required.
3. Select a Vendor. Supported vendors are OpenAI, Claude, and AWS Bedrock.
4. Enter a Vendor Model ID. The Vendor Model ID identifies the specific model version from the selected vendor.
5. Optionally enter a Description.
6. Set the three generation parameters. These are standard LLM controls, so they will be familiar from the vendor's own documentation. Each has a slider and a reset control. Use low Temperature and Top P values when you want strict accuracy and logical, consistent output, such as for extracting data or generating code. Use high values when you want creative, varied output. Adjust one of the two rather than both. Some models do not support these parameters. Newer Claude models reject Temperature and Top P, and setting them either has no effect or causes an error. Check your vendor's documentation to confirm which parameters the model you are connecting to accepts.
7. 1. Maximum Tokens: The limit on how many tokens the model can produce in a response, which caps response length.
   2. Temperature: Controls how predictable the model's output is. A low value keeps the model focused on its most likely responses, which produces consistent, repeatable output. A high value produces more varied and creative output.
   3. Top P: Controls how many candidate words the model considers at each step. A low value restricts the model to its most probable choices. A high value allows a wider range of possibilities.
8. Configure the Secrets section. These are the credentials that allow the connection to authenticate with the vendor, and the required fields differ by vendor.
9. 1. OpenAI: Project ID, API Key, and Organization ID.
   2. Claude: API Key.
   3. AWS Bedrock: Role ARN.
10. Optionally configure the Pricing configuration section. Both fields are optional and read No value set until you enter a price. Each has a slider and a numeric input, so you can drag the slider or type an exact value, and each has a Restore button.
11. 1. Input Cost: The price in USD charged per one million input (prompt) tokens sent to the model.
    2. Output Cost: The price in USD charged per one million output (completion) tokens returned by the model.
12. Click Create to complete the form.

If required credentials have not been configured, a Secrets missing warning appears on the model card in the list view.

Model Detail Page

After creation, the model detail page displays the vendor, Vendor Model ID, parameter sliders, masked secrets fields, and, when pricing has been configured, the Input Cost and Output Cost values. Each field has a Restore button.

An overflow menu in the top right provides Collapse and Delete options.