Agent Configuration via `agents.yaml`¶

The AI-Parrot AgentRegistry allows you to define and manage agents declaratively using a YAML configuration file. This approach is preferred for static agent definitions, enabling easy modification of models, tools, and behaviors without changing code.

File Location¶

The configuration file is located at: agents/agents.yaml (relative to your project root).

Basic Structure¶

The file expects a top-level agents key containing a list of agent definitions.

agents:
  - name: "MyAgent"
    class_name: "BasicAgent"
    module: "parrot.bots.agent"
    enabled: true
    # ... further configuration

Configuration Reference¶

Core Fields¶

Field	Type	Required	Description
`name`	string	Yes	Unique identifier for the agent. Used to retrieve instances.
`class_name`	string	Yes	The class name of the bot (e.g., `BasicAgent`, `Chatbot`).
`module`	string	Yes	The python module path where the class is defined.
`enabled`	boolean	No	Defaults to `true`. Set to `false` to disable loading.
`singleton`	boolean	No	If `true`, only one instance of this bot is created.
`at_startup`	boolean	No	If `true`, the bot is instantiated immediately when the registry loads.

Model Configuration¶

You can define the LLM in two ways:

1. Simple String Format: client:model_name

model: "openai:gpt-4o"

2. Detailed Dictionary

model:
  client: "anthropic"
  model: "claude-3-5-sonnet-20240620"
  temperature: 0.7  # Optional parameters passed to config

System Prompt¶

Directly define the system instructions for the agent.

system_prompt: |
  You are a helpful coding assistant.
  Always answer in Python.

Tools¶

List of tool names (strings) available in the ToolManager.

tools:
  - "google_search"
  - "calculator"

Toolkits¶

List of toolkit names to register.

toolkits:
  - "data_analysis_toolkit"

MCP Servers (Model Context Protocol)¶

Connect to MCP servers to dynamically load tools.

mcp_servers:
  - name: "filesystem"
    transport: "stdio"
    command: "npx"
    args: ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/allowed_dir"]

  - name: "weather_service"
    transport: "sse"
    url: "http://localhost:8080/sse"

Attributes: - name: Identifier for the server. - transport: stdio (default) or sse (HTTP). - command: Executable to run (for stdio). - args: List of arguments for the command. - url: URL endpoint (for sse). - env: Dictionary of environment variables.

Vector Store (Memory)¶

Configure vector memory for the agent.

vector_store:
  vector_store: "pgvector"  # or 'chroma', 'qdrant', etc.
  collection_name: "agent_memory"
  dimension: 1536
  embedding_model: "openai"

Additional Configuration (`config`)¶

Any extra key-value pairs in the config block are passed directly to the Agent's __init__ method as keyword arguments.

config:
  verbose: true
  max_history: 10
  user_id: "default_user"

Complete Example¶

agents:
  - name: "ResearchBot"
    class_name: "BasicAgent"
    module: "parrot.bots.agent"
    enabled: true
    description: "An agent that researches topics using Google and MCP."

    # Model Setup
    model:
      client: "openai"
      model: "gpt-4-turbo"

    # Behavior
    system_prompt: |
      You are a senior researcher.
      Use Google Search for latest info and Filesystem to save reports.

    # Capabilities
    tools:
      - "google_search"

    # MCP Integration
    mcp_servers:
      - name: "filesystem"
        command: "npx"
        args: ["-y", "@modelcontextprotocol/server-filesystem", "./research_reports"]

    # specific arguments for BasicAgent
    config:
      speech_context: "Formal"
      report_template: "research_layout.html"

Agent Configuration via agents.yaml¶