Skip to main content

Models

Create a Model

Creates a new custom model for the workspace. Provider credentials in the configuration are encrypted using the workspace encryption key before being persisted.

Create Autorouter

Creates an autorouter model that routes between a strong and economical source model based on the requested profile. Both source models must already exist for the workspace and be marked autorouter-eligible in master data.

Update Autorouter

Re-configures an autorouter model. Each of key/strong_model/economical_model/profile falls back to the existing value when omitted. Changing the key enforces uniqueness and rewrites PRICING_KV.

Create AWS Bedrock

Registers an AWS Bedrock inference profile as a custom model for the workspace. Credentials are resolved at request time via either the integration reference or pod-identity: nothing is stored with the model.

Validate AWS Bedrock

Performs a live Bedrock Converse probe to verify the inference profile ARN and credentials, then best-effort enriches the response from known system models.

Update AWS Bedrock

Updates an AWS Bedrock custom model. ARN changes are format-validated (live AWS validation lives in the dedicated validate endpoint). Configuration and metadata are spread-merged. Parameters are replaced only when the request produces a non-empty list.

Azure Foundry Deployments

Lists Azure Foundry deployments under the given base_url and joins each entry with the Orq master-data row. Only OpenAI-developed deployments in succeeded state with chat/completion/embedding/vision model types are returned.

Import LiteLLM

Bulk-imports a list of LiteLLM model definitions into the workspace model garden.

List LiteLLM

Fetches the list of models from the LiteLLM instance configured for the workspace. Requires a stored LiteLLM integration.

Create OpenAI Like

Creates a custom model backed by any OpenAI-compatible endpoint. The handler probes the target API with the supplied credentials before persisting the model.

Update OpenAI Like

Updates an OpenAI-compatible custom model. Live-re-probes the target API when base_url or model_id changes, using the stored encrypted api_key. Metadata is merged (existing preserved, new overrides).

Validate a Model

Validates a provider endpoint by performing a minimal live probe. Currently supports Azure OpenAI. Response includes the resolved region, whether the model is known to Orq, and either the full model document or a synthesized default.

Create Vertex

Registers a Google Vertex AI model as a custom model for the workspace. The service account credentials are probed against Vertex AI with a minimal GenerateContent call before persisting.

Delete a Model

Deletes a custom model from the workspace. System models cannot be deleted. Returns 200 with an explanatory message if the model is a system model or is still referenced by experiments.

Update a Model

Updates a custom model. Only fields present in the request body are modified, except for metadata and parameters, which are fully replaced when present (preserved from the legacy handler’s behavior).

Enable a Model

Adds the model to the workspace’s enabled set. Idempotent: re-enabling an already-enabled model returns 204 with no state change.

Disable a Model

Removes the model from the workspace’s enabled set. Idempotent: disabling an already-disabled model returns 204.

List Models

Lists all models available through the AI Router. Returns each model in OpenAI-compatible shape with its provider, ID, and creation timestamp.