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
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