Custom Models

Open My Account → Custom Models to add custom LLM connections that every project on this team can use.

Models you add here are team-scoped. They show up for agents in all projects on the team. Project-only custom models are still configured on the agent; see Predefined and Custom Models.

The endpoint is validated before the model is saved.

Informação

Only team admins can open this tab and add, edit, or delete team custom models.

Empty State

If the team has no custom models yet, the page shows No custom models yet and Add Custom Model.

Add a Custom Model

1. Click Add Custom Model

   The New Connection dialog opens.

2. Fill in the Fields

   

   Field	Description
   Model Name	Identifier the API expects (for example gpt-4o)
   Model Provider	Provider from the dropdown (for example Open AI)
   Configure through kong gateway?	Optional. Route the connection through the Kong gateway.
   API Key	Provider key. Stored securely.
   API URL	API endpoint
   Timeout (seconds)	Default is 300. Increase for long-running reasoning models.
   Ignore temperature (creativity)	Some newer models reject requests that send a temperature/creativity parameter. Enable this so temperature is not sent for this model, regardless of provider.
   Parse thinking tags (<think>/</think>)	Some reasoning models inline a chain of thought as <think>…</think> in the message content (for example a LiteLLM proxy with merge_reasoning_content_in_choices enabled). Enable this to strip those tags from the answer and show the reasoning separately.
   Max Input Tokens	Default 3000
   Max Response Tokens	Default 1000
   Thinking	Enable provider thinking/reasoning for this model.
   Thinking Budget	Token budget for thinking, when Thinking is on.
   Include Thoughts	Include the model's thoughts in the response.

   Other provider-specific fields (Azure API version, AWS timeout, custom auth headers, dynamic token fetching) appear when that provider is selected.

3. Create

   Click Create. EKB validates the endpoint; on success the model is stored for the team. Cancel closes the dialog without saving.

Model List

When at least one model exists, the page shows a search box (Search by name or provider...) and a table:

Column	Description
Name	Model name
Provider	Provider label
Max Input	Max input tokens
Max Response	Max response tokens
Actions	Edit or delete

If search matches nothing: No models match your search.

Edit or Delete

• Edit reopens the same form with the current values.
• Delete asks for confirmation, then removes the team model.

Team models cannot be edited from a project's custom-model list. There they appear as read-only with Managed in team settings.

Related

• Predefined and Custom Models — catalog of built-in models and project-level BYOK
• Agent Mode