Skip to content

Configuration

Genie Tooling is configured at runtime using a MiddlewareConfig object. For ease of use, especially for common setups, MiddlewareConfig integrates a FeatureSettings model.

Simplified Configuration with FeatureSettings

The recommended way to start configuring Genie is by using the features attribute of MiddlewareConfig. FeatureSettings provides high-level toggles and default choices for major components like LLM providers, RAG components, caching, tool lookup, logging adapter, observability, HITL, token usage, guardrails, prompt system, conversation state, and distributed task queues.

from genie_tooling.config.models import MiddlewareConfig
from genie_tooling.config.features import FeatureSettings

app_config = MiddlewareConfig(
    features=FeatureSettings(
        llm="ollama",
        command_processor="llm_assisted",
        # ... other features ...
    )
)

How FeatureSettings Works

When you initialize Genie with a MiddlewareConfig containing FeatureSettings, an internal ConfigResolver processes these settings. It translates your high-level choices into specific plugin IDs and default configurations for those plugins. For a full list of available aliases and more details on simplified configuration, please see the Simplified Configuration Guide.

Explicit Overrides and Detailed Configuration

While FeatureSettings provides a convenient starting point, you can always provide more detailed, explicit configurations that will override or augment the settings derived from features.

Explicit Tool Enablement: auto_enable_registered_tools and tool_configurations

This is a critical security and configuration concept in Genie Tooling.

  • auto_enable_registered_tools: bool (Default: True)

    • When True (for development), tools registered via the @tool decorator and genie.register_tool_functions() are automatically enabled. Class-based tools still generally need to be listed in tool_configurations to be enabled.
    • When False (for production), a tool is only active if its identifier is explicitly listed as a key in the tool_configurations dictionary. This applies to both class-based plugins and @tool decorated functions.
    • Recommendation: For production environments, it is strongly recommended to set auto_enable_registered_tools=False to maintain a clear, secure manifest of the agent's capabilities.

  • tool_configurations: Dict[str, Dict[str, Any]]

    • This dictionary serves two purposes:
      1. Enabling Tools: If a tool's identifier is a key in this dictionary, it is considered enabled.
      2. Configuring Tools: The value associated with the key is a dictionary passed to the tool's setup() method. Use an empty dictionary {} to enable a tool that requires no configuration.

Production Example: 

app_config = MiddlewareConfig(
    auto_enable_registered_tools=False, # Production-safe setting
    features=FeatureSettings(
        llm="openai",
        llm_openai_model_name="gpt-4-turbo-preview",
    ),
    # Explicitly enable and configure tools
    tool_configurations={
        "calculator_tool": {}, # Enable calculator with no config
        "sandboxed_fs_tool_v1": {"sandbox_base_path": "./agent_workspace"},
        "my_registered_tool_function": {} # Enable a decorated tool
    },
    # ... other specific configurations ...
)

Configuring Other Plugins

You can provide specific configurations for individual plugins using the various *_configurations dictionaries in MiddlewareConfig. These dictionaries are keyed by the canonical plugin ID or a recognized alias.

app_config = MiddlewareConfig(
    features=FeatureSettings(llm="openai"),
    llm_provider_configurations={
        "openai_llm_provider_v1": { # Canonical ID
            "model_name": "gpt-4o", # Overrides model from features
            "request_timeout_seconds": 120
        }
    },
    # ...
)

Plugin Development Directories

If you have custom plugins located outside your main Python path or not installed as entry points, you can specify their location:

app_config = MiddlewareConfig(
    plugin_dev_dirs=["/path/to/my/custom_plugins", "./project_plugins"]
)
The PluginManager will scan these directories for valid plugin classes.