Agent Configuration

Master the blueprint of your agentic network by understanding the core configuration fields and validation rules.

This guide provides a comprehensive reference for every field available when defining an agent in SVAHNAR, along with the strict logic rules required for a successful deployment.

💡 Core Concepts

The configuration of an agent defines its identity, intelligence, and connectivity. It tells the system who the agent is (Name/Role), how it thinks (LLM Config), and where it fits in the network (Edges).

Key Components:

• Identity: agent_name and Role.
• Intelligence: LLM_config and agent_function.
• Connectivity: incoming_edge and outgoing_edge.

---

⚙️ Field Reference

Agent Fields

These are the top-level properties required to define an agent.

Field	Type	Required	Description	Example
agent_name	string	Yes	A unique identifier for the agent within the network.	"research_agent"
LLM_config	array	Yes	Configuration settings for the AI model driving this agent.	(See LLM Configurations below)
Role	string	Yes (for Supervisor Agents)	Defines if the agent acts as a manager. Values: normal or supervisor.	"supervisor"
agent_function	string	Yes	The prompt or system instruction that defines the agent's goal.	"Your goal is to extract email addresses..."
incoming_edge	array	Yes	A list of agents (or "Start") that trigger this agent.	["Start"]
outgoing_edge	array	Yes (if exists)	A list of agents that this agent triggers next.	["writer_agent"]
tools	object	Yes (for non-supervisor/normal agents)	External capabilities assigned to the agent.	See Tool Configuration

LLM Configurations

The LLM_config block dictates the "brain" of the agent.

Enterprise Feature

The provider field (BYOAI - Bring Your Own AI) is strictly for Enterprise users. Standard users should omit this field and use the default SVAHNAR models.

Field	Type	Required	Description	Example
provider	string	No*	The AI provider (e.g., ChatOpenAI, ChatNvidia). Enterprise only.	"ChatOpenAI"
params	object	Yes	Specific parameters for the chosen model.	(See Parameters below)

Parameters Object:

Field	Type	Required	Description	Allowed Values
model	string	Yes	The specific model version to use.	"gpt-5", "gpt-5-mini"
max_tokens	int	No	The maximum length of the generated response.	(e.g., 4000)
request_timeout	int	No	The timeout limit in seconds.	(e.g., 600)

---

🛑 Validation Rules

To ensure your network deploys without errors, your configuration must adhere to these rules.

Note

If you are creating an agent using the SVAHNAR SDK, you can use the validate function to validate your agent configuration. If you're using the Agent Console, this validation is performed automatically.

1. Structure & Naming

• Top-Level Key: The YAML file must start with create_agent_network.
• Uniqueness: Every agent_name must be unique across the entire file.

2. Edges & Connectivity

• The Start Node:
  • Exactly one agent must have "Start" in its incoming_edge.
• The End Node:
  • outgoing_edge must never contain "Start".
• Validity: All agents listed in edges must exist in the agent_name definitions.
  • All agent names must be in camelCase or snake_case.

3. Role Constraints

• Supervisors: An agent with Role: "supervisor" cannot have tools assigned to it.

4. Model Compliance

• Ensure your LLM_config parameters match the allowed model list (e.g., gpt-5).

---

📚 Practical Recipes (Example)

Recipe: Standard Agent Configuration

Use Case: A standard agent connected to the start node, using the GPT-5 Mini model. Check single agent configuration and multi agent configuration