# Governance types
Syntphony CAI provides multiple Governance models that define how interactions are interpreted, routed, and executed.
These models exist along a continuum of **autonomy and control**, enabling the selection of an appropriate balance between deterministic behavior and dynamic execution.
The choice of model depends on:
* Use case complexity;
* Predictability of user interactions;
* Required level of control.
## Models overview:
From left to right, Governance models increase in **flexibility** and **autonomy**:
* **Intent-based (NLU):** Fully structured and deterministic;
* **Composite (NLU-first):** Structured with controlled access to Agents;
* **Composite (Agentic-first):** Agent-driven with selective structure; and
* **Agentic:** Fully dynamic, Agent-based execution.
Governance models
## Intent-based model (NLU)
The Intent-based model (NLU) is designed for structured interactions with **predefined conversational paths,** where:
* User input is classified into intents;
* Each intent maps to a predefined process or response.
This ensures high **predictability** and **consistency**, making it suitable for well-defined and repeatable scenarios.
#### **Typical use cases:**
* Order tracking;
* FAQ handling;
* Basic customer support; and
* Information retrieval.
{% hint style="info" %}
This model provides strong control and reliability, with minimal variability in outcomes.
{% endhint %}
## Agentic model
The Agentic model enables **dynamic**, **context-driven execution** through AI Agents coordinated by the Supervisor, where:
* Requests are evaluated and routed dynamically;
* Agents execute tasks using Actions, Tools, and structured capabilities; and
* Planning and execution can adapt based on context.
#### This governance model is suited for:
* Complex problem-solving;
* Multi-step workflows; and
* High-variability scenarios.
{% hint style="info" %}
Despite its flexibility, execution remains governed through defined rules, capabilities, and orchestration.
{% endhint %}
## Composite (Hybrid) model
The Composite model combines structured intent-based routing with Agent-based execution.
**It allows both deterministic and dynamic handling** within the same system, enabling a controlled transition between models.
### NLU-first
In this configuration, **intent recognition is the primary entry point, where:**
* User input is first matched against predefined intents;
* If a match **is found:** Then it will be handled through predefined flows;
* If **no match** is found: Then the system can transfer execution to an Agent.
Agents are not directly selectable in this model. They can only be invoked through an explicit [**Transfer cell** ](https://docs.conversational-ai.syntphony.com/user-guide/build-dialogs/dialog-cells/transfer) within an NLU flow.
The Transfer cell is executed after intent classification, enabling controlled delegation from structured flows to Agent-based execution, as we see on the example below:
Intent and transfer cells on a user journey flow
The NLU-first model is ideal when:
* Most interactions are predictable;
* Agent-based handling is required only for exceptions.
{% hint style="info" %}
This configuration is also suitable for gradually evolving NLU-based systems toward Agentic architectures.
{% endhint %}
### Agentics-first
In this configuration, **Agent-based execution is the primary entry point**, where:
* Requests are first handled by the Supervisor and routed to Agents;
* Intent-based routing is used as a secondary mechanism for specific or deterministic cases.
The Agentics-first model is best suited when:
* Interactions are diverse and less predictable;
* Flexibility is prioritized over strict structure.
{% hint style="info" %}
It enables dynamic execution while preserving selective control through intent-based logic.
{% endhint %}
## Models compatibility
After creating a Project, certain restrictions may apply when changing the governance model, depending on the components and configurations already in use.
