Skip to content

Individual agent configurations define the behavior, capabilities, and settings for each agent in your manifest. Each agent entry in the agents dictionary represents a complete agent setup.

Overview

Agent configuration includes:

  • Model settings: LLM provider and model selection
  • System prompts: Define agent behavior and personality
  • Tools and toolsets: Capabilities available to the agent
  • Knowledge sources: Context and information access
  • Output types: Structured response definitions
  • Workers: Sub-agents for delegation
  • Connections: Message routing to other nodes
  • MCP servers: Model Context Protocol integrations
  • Triggers: Event-based activation

Configuration Reference

Agent Configuration

Configuration for a single agent in the system.

Defines an agent's complete configuration including its model, environment, and behavior settings.

Docs: https://phil65.github.io/llmling-agent/YAML%20Configuration/agent_configuration/

Agent Configuration (YAML)
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
inherits: null  # Name of agent config to inherit from
model: null  # The model to use for this agent. Can be either a simple model name
tools: []  # A list of tools to register with this agent.
toolsets: []  # Toolset configurations for extensible tool collections.
session: null  # Session configuration for conversation recovery.
output_type: null  # Name of the response definition to use.
retries: 1  # Number of retries for failed operations (maps to pydantic-ai's retries)
output_retries: null  # Max retries for result validation
end_strategy: early  # The strategy for handling multiple tool calls when a final result is found
avatar: null  # URL or path to agent's avatar image
system_prompts: []  # System prompts for the agent. Can be strings or structured prompt configs.
knowledge: null  # Knowledge sources for this agent.
workers: []  # Worker agents which will be available as tools.
requires_tool_confirmation: per_tool  # How to handle tool confirmation:
debug: false  # Enable debug output for this agent.
environment: null  # Execution environment configuration for this agent.
usage_limits: null  # Usage limits for this agent.
tool_mode: null  # Tool execution mode:
auto_cache: 'off'  # Automatic prompt caching configuration:
name: null  # Identifier for the node. Set from dict key, not from YAML.
config_file_path: null  # Config file path for resolving relative paths.
display_name: null  # Human-readable display name for the node.
description: null  # Optional description of the agent / team.
triggers: []  # Event sources that activate this agent / team
connections: []  # Targets to forward results to.
mcp_servers: []  # List of MCP server configurations:
input_provider: null  # Provider for human-input-handling.
event_handlers: []  # Event handlers for processing agent stream events.

Configuration Inheritance

Agents can inherit configuration from other agents or base configurations:

agents:
  base_agent:
    model: "openai:gpt-4o"
    retries: 2
    toolsets:
      - type: "resource_access"

  specialized_agent:
    inherits: "base_agent"
    description: "Specialized version"
    system_prompts:
      - "You are a specialized agent..."

Agents Section

Complete example of an agent configuration:

agents:
  web_assistant:                   # Name of the agent
    description: "Helps with web tasks"  # Optional description
    model: openai:gpt-5           # Model to use
    tools:
      open_browser:
        import_path: webbrowser.open
        description: "Opens URLs in browser"
    system_prompts:
      - "You are a web assistant."
      - "Use open_browser to open URLs."
    retries: 2                   # Number of retries for failed

Field Reference

Field Name Description
name Identifier for the agent (set from dict key, not from YAML)
config_file_path Config file path for resolving relative paths
display_name Human-readable display name for the agent
description Optional description of the agent
triggers Event sources that activate this agent
connections Targets to forward results to
mcp_servers List of MCP server configurations
input_provider Provider for human-input-handling
event_handlers Event handlers for processing agent stream events
inherits Name of agent config to inherit from
model The model to use for this agent
tools A list of tools to register with this agent
toolsets Toolset configurations for extensible tool collections
session Session configuration for conversation recovery
output_type Name of the response definition to use
retries Number of retries for failed operations
output_retries Max retries for result validation
end_strategy The strategy for handling multiple tool calls when a final result is found
avatar URL or path to agent's avatar image
system_prompts System prompts for the agent
knowledge Knowledge sources for this agent
workers Worker agents which will be available as tools
requires_tool_confirmation How to handle tool confirmation (always/never/per_tool)
debug Enable debug output for this agent
environment Execution environment configuration for this agent
usage_limits Usage limits for this agent
tool_mode Tool execution mode (None/codemode)
auto_cache Automatic prompt caching configuration