Every AI feature in SESTEK AI Agents reaches a large language model through an LLM Configuration that belongs to your tenant. A configuration holds a provider and its credentials. The model is chosen where the configuration is used, so one configuration can serve an AI agent's chat model and a knowledge base's embedding model.
This page shows where each feature chooses its LLM. Creating a configuration, the providers and their fields, and choosing a model are covered in Creating an LLM Configuration.
📝 Note: No AI feature can be set up until the tenant has an LLM Configuration for the provider you want. Create it under Integrations > LLM Configurations first, then choose it in the feature.
Where each feature chooses its LLM
| Feature | Where you choose the LLM |
|---|---|
| AI agents | Each agent's settings: LLM Configuration and Model. |
| Ask LLM node | On the node: LLM Configuration (required) and Model. |
| RAG Agent node | On the node: LLM Configuration (required) and Model. |
| Advanced Conversation Routing | The Virtual Agent node's LLM Configuration and Model, with LLM Request Timeout for ACR (sec). Default: 3 (range 1-10). If the model does not answer in time, the NLU engine's intent is used. |
| Agent Builder and Improver | Administration > Settings > Agent Builder & Improver |
| AI Testing (simulated conversations, evaluation, scenario generation, optimization) | Administration > Settings > AI Testing |
| Custom Dashboards (chart assistant) | Administration > Settings > Custom Dashboard |
| AI Insights | Administration > Settings > Add-On AI Features |
| Add-On AI Features | Administration > Settings > Add-On AI Features |
| Guardrails | Content filters: the guardrail's Content Filters step. A block message in LLM mode is written with the AI agent's own model and provider settings, not the content filter's configuration. |
| Knowledge bases | Each knowledge base: Embedding Configurations (required), Document Processing and Retrieval Strategy (optional). Reranking, under Retrieval Strategy, works only with an Azure OpenAI or Sestek configuration. |
📝 Note: A knowledge base's embedding configuration and its Token Per Minute cannot be changed after the knowledge base is created. Choose them before you create it.
The four tabs under Administration > Settings in the table (Add-On AI Features, AI Testing, Agent Builder & Improver and Custom Dashboard) set one LLM for the whole tenant. Each holds the same three fields, LLM Configuration, Model and Reasoning Effort, described in Creating an LLM Configuration. You need permission to manage settings to open them.
Changing models across a project
Project Settings > Model Assignments changes the model of a project's AI agents and its Ask LLM, RAG Agent and Virtual Agent nodes in one place, but not of its knowledge bases or guardrails. You can edit it only on the draft version, and it is read-only on a published one. See Model Assignments.
Where the calls are made, and what reaches the model
SESTEK's AI module services call the LLM provider. The admin panel and the conversation runtime do not. In an on-premises installation, the AI module services need outbound network access to the endpoint of every LLM provider you configure.
When a guardrail with PII masking is attached to an AI agent, the user's message is masked before it reaches the agent's model.
Next
- Creating an LLM Configuration: creating, editing and deleting a configuration, the fields for each provider, and choosing a model and Reasoning Effort where the configuration is used.
- Add-On AI Features: the smaller AI helpers that share the Add-On AI Features setting, and where each one appears.