# Reference for the teleport\_inference\_model Terraform resource

This page describes the supported values of the teleport\_inference\_model resource of the Teleport Terraform provider.

## Schema

### Required

- `metadata` (Attributes) (see [below for nested schema](#nested-schema-for-metadata))
- `spec` (Attributes) (see [below for nested schema](#nested-schema-for-spec))

### Optional

- `sub_kind` (String) SubKind is the resource sub-kind. Should be empty.
- `version` (String) Version is the resource version. Should be set to "v1".

### Nested Schema for `metadata`

Required:

- `name` (String) name is an object name.

Optional:

- `description` (String) description is object description.
- `expires` (String) expires is a global expiry time header can be set on any resource in the system.
- `labels` (Map of String) labels is a set of labels.

### Nested Schema for `spec`

Optional:

- `bedrock` (Attributes) Bedrock indicates that this model uses Amazon Bedrock as the inference provider and specifies Bedrock-specific parameters. (see [below for nested schema](#nested-schema-for-specbedrock))
- `max_session_length_bytes` (Number) MaxSessionLengthBytes is the maximum session length that can be sent to inference provider. Currently, it's determined by the size of model's context window; future versions of Teleport will allow summarizing larger sessions by splitting them. Inference providers will reject requests that are larger than given model's context window. Since context windows are usually sized in tokens, this value is an approximation. Assuming 2 bytes per input token should be safe. Currently, Teleport will outright reject sessions larger than this limit; future versions will split sessions in chunks, treating this size as a maximum. If unset or set to 0, defaults to 1MB.
- `openai` (Attributes) Openai indicates that this model uses OpenAI as the inference provider and specifies OpenAI-specific parameters. (see [below for nested schema](#nested-schema-for-specopenai))

### Nested Schema for `spec.bedrock`

Optional:

- `bedrock_model_id` (String) BedrockModelId specifies a model ID or an inference profile as understood by the Bedrock API.
- `integration` (String) Integration is the AWS OIDC Integration name. If unset, Teleport will use AWS credentials available on the Auth Service machine; otherwise, it will use the specified OIDC integration for assuming appropriate role.
- `region` (String) Region is the AWS region which will be used for inference.

### Nested Schema for `spec.openai`

Optional:

- `api_key_secret_ref` (String) ApiKeySecretRef is a reference to an InferenceSecret that contains the OpenAI API key.
- `base_url` (String) BaseUrl is the OpenAI API base URL. Optional, defaults to the public OpenAI API URL. May be used to point to a custom OpenAI-compatible API, such as LiteLLM. In such case, the `api_key_secret_ref` must point to a secret that contains the API key for that custom API.
- `openai_model_id` (String) OpenaiModelId specifies the model ID, as understood by the OpenAI API.