This article covers creating, editing and deleting an LLM Configuration, and choosing a model and Reasoning Effort wherever the configuration is used. What a configuration is and where each feature chooses one are covered in LLM Integrations.
To create, edit or delete a configuration you need permission to edit LLM Configurations. Without it, a configuration opens read-only as LLM Configuration Details, with no Save button.
Creating a configuration
Configurations belong to the whole tenant. Go to Integrations > LLM Configurations and click New LLM Configuration. Choose a provider, then fill in the fields that provider needs. Every field is required when you create a configuration, and keys and secrets are masked as you type.
The example below connects an Azure OpenAI deployment called support-chat.
1. Name: Unique in the tenant. For the example, azure-openai-prod.
2. Provider: The service the configuration connects to. The fields below it change with the provider. For the example, Azure OpenAI.
3. API Key: The key of your Azure OpenAI resource.
4. API Base: The resource's endpoint, in the form https://{resource_name}.openai.azure.com. For the example, https://contoso-support.openai.azure.com.
5. Deployment Name: The deployment to call, here support-chat. For Azure OpenAI the deployment decides the model, so you do not choose a model anywhere else.
6. API Version: The API version to use when calling the deployment.
Fields for each provider
| Provider | Fields |
|---|---|
| OpenAI, Google Gemini, Groq, Together AI | API Key |
| Azure OpenAI | API Key, API Base (for example https://{resource_name}.openai.azure.com), Deployment Name, API Version |
| AWS Bedrock | AWS Bedrock Region, AWS Bedrock Access Key ID, AWS Bedrock Secret Access Key |
| Google Vertex AI | Vertex AI Project ID, Vertex AI Location, Vertex AI Private Key, Vertex AI Client Email, Vertex AI Token URI |
| Sestek | API Key, API Base: an OpenAI-compatible model server (vLLM) at the address you give |
| SESTEK Proxy | API Key, API Base (for example https://your-proxy.example.com/): an OpenAI-compatible gateway such as a LiteLLM proxy |
Editing a configuration
Open a configuration from the LLM Configurations list to edit it.
| Item | When you edit |
|---|---|
| Name | Editable |
| Provider | Fixed. To use another provider, create a new configuration. |
| Keys and secrets | Shown empty, with the hint Enter to update current value. Leave a field empty to keep the stored value, or type to replace it. |
| Other fields | Editable |
📝 Note: Saving a configuration moves every AI agent that uses it to a state where the agent must be prepared again. See The AI Agent Lifecycle.
Choosing a model where the configuration is used
The same LLM Settings fields appear wherever a configuration is used: in an AI agent, on a node, or on a settings tab such as the one for Add-On AI Features.
1. LLM Configuration: The tenant's configurations. Choose the one this feature should use.
2. Model: The models suggested for that configuration's provider. Models marked legacy are no longer listed. If the model you want is not listed, choose custom model and type its name in Model Name.
3. Reasoning Effort: None, Low, Medium, High or Default. For a reasoning model chosen from the list, it starts at None. None turns reasoning off, or sets it to the lowest level for a model that cannot switch reasoning off. Default sends no value and leaves the choice to the provider.
The provider decides how the Model and Reasoning Effort fields behave:
| Provider | Model | Reasoning Effort |
|---|---|---|
| OpenAI, AWS Bedrock, Google Gemini, Google Vertex AI, Groq, Together AI | Choose from the suggested list, or choose custom model and type the name in Model Name | For a model from the list, shown only if the model supports reasoning, and then it starts at None. For a custom model, it is off until you turn on its Enable switch. |
| Sestek, SESTEK Proxy | No list: choose custom model and type the name in Model Name | Off until you turn on its Enable switch |
| Azure OpenAI | No Model field: the Deployment Name decides the model | Not available |
⚠️ Warning: For a custom model, turn on Enable only if that model supports reasoning. On a model that does not, the agent's call to the model fails.
Embedding models
A knowledge base chooses its embedding model in the same way. Suggested embedding models exist for OpenAI, Azure OpenAI, Google Gemini and Google Vertex AI. custom model is offered for every provider.
Deleting a configuration
A configuration cannot be deleted while either of these is true:
- It is selected on one of the Add-On AI Features, AI Testing, Agent Builder & Improver or Custom Dashboard tabs under Administration > Settings.
- An AI agent, guardrail, knowledge base or flow uses it on a draft or published project version.
Choose another configuration in those places first. If only older project versions use it, you are warned that ongoing sessions on those versions may be affected, and you can confirm the delete.
Limits and naming rules
| Item | Value |
|---|---|
| Name | Up to 200 characters: letters, digits, underscores and hyphens. Must be unique in the tenant. |
| Other fields | Long enough for any real value |