Understand Agent Fields and Validation Rules

Here's a table summarizing the main fields for each agent:

Agent Fields

Field	Type	Description	Required	Example
agent_name	string	Unique name for the agent	Yes	"example_agent"
LLM_config	array	Configuration of the LLM(s) used by the agent	Yes	See LLM Configurations
Role	string	Role of the agent (normal or supervisor)	No	"supervisor"
agent_function	string	List of functions the agent performs	Yes	"Your goal is to send a request to the API endpoint with the given parameters. Make sure you check the parameters before sending the request. Parameters are: "param1": "value1", "param2": "value2"
incoming_edge	array	List of agents or Start connected to this agent	Yes	["Start"]
outgoing_edge	array	List of agents connected from this agent	Yes	["agent-2"]
tools	object	Tools assigned to the agent	No	See Tool Configuration

LLM Configurations

Each entry in LLM_config includes:

Field	Type	Description	Required	Example
provider	string	LLM provider (ChatOpenAI, ChatNvidia)	No - Only for Enterprise (BYOAI)	"ChatOpenAI"
params	object	Provider-specific parameters	Yes	See below

BYOAI (Bring Your Own AI) is only available for Enterprise users.

Parameters:

Field	Type	Allowed Values	Required	Example
model	string	"gpt-5", "gpt-5-mini"	Yes	"gpt-5-mini"
max_tokens	integer	Minimum: 1	No	500
request_timeout	integer	Minimum: 1	No	30

Validation Rules

• Top-Level Key: Must only contain create_vertical_agent_network.
• Agent Names: Each agent's agent_name must be unique.
• Edges:
  • incoming_edge can contain Start (only once across all agents).
  • outgoing_edge must not contain Start.
  • Edges must reference valid agent names.
• Supervisor Limitations:
  • supervisor agents can only connect to agents defined earlier in the YAML file.
• LLMs:
  • Ensure and params matches the allowed models.
• Start Node:
  • Exactly one agent must have Start in its incoming_edge.