PresetManager

The PresetManager class provides a singleton-based management system for AI model presets.

Overview

PresetManager is the recommended way to manage model presets in AmritaCore. Instead of manually creating and handling ModelPreset instances, you should use PresetManager to centralize preset management, ensuring consistency and reducing configuration errors.

When no preset is explicitly selected, the system will automatically fallback to a default preset. This fallback mechanism ensures your application continues to work even if preset selection fails or is not specified.

Properties

• _default_preset (ModelPreset | None): The default preset to use when none is specified
• _presets (dict[str, ModelPreset]): Internal storage for all registered presets

Methods

__new__() -> Self

Creates or returns the singleton instance of PresetManager.

__init__() -> None

Initializes the PresetManager (only runs once due to singleton pattern).

set_default_preset(preset: ModelPreset | str) -> None

Sets the default preset to use when no specific preset is selected.

Parameters:

• preset: Either a ModelPreset object or the name of an existing preset

Example:

from amrita_core.preset import PresetManager
from amrita_core.types import ModelPreset

manager = PresetManager()

# Set using ModelPreset object
preset = ModelPreset(model="gpt-3.5-turbo", api_key="your-key")
manager.set_default_preset(preset)

# Or set using preset name
manager.set_default_preset("my-preset-name")

get_default_preset() -> ModelPreset

Returns the default preset. If no default has been set, it will automatically select a random preset from available presets.

Returns:

• ModelPreset: The default preset configuration

Example:

manager = PresetManager()
default = manager.get_default_preset()
print(f"Default preset: {default.name}")

get_preset(name: str) -> ModelPreset

Retrieves a specific preset by name.

Parameters:

• name: The identifier name of the preset

Returns:

• ModelPreset: The requested preset configuration

Raises:

• ValueError: If the preset name doesn't exist

Example:

try:
    preset = manager.get_preset("gpt-4-preset")
except ValueError as e:
    print(f"Preset not found: {e}")

add_preset(preset: ModelPreset) -> None

Adds a new preset to the manager.

Parameters:

• preset: The ModelPreset object to register

Raises:

• ValueError: If a preset with the same name already exists

Example:

preset1 = ModelPreset(model="gpt-3.5-turbo", name="fast-model", api_key="your-key")
preset2 = ModelPreset(model="gpt-4", name="smart-model", api_key="your-key")

manager.add_preset(preset1)
manager.add_preset(preset2)

get_all_presets() -> list[ModelPreset]

Returns all registered presets.

Returns:

• list[ModelPreset]: A list of all preset configurations

Example:

all_presets = manager.get_all_presets()
for preset in all_presets:
    print(f"- {preset.name}: {preset.model}")

async test_single_preset(preset: ModelPreset | str) -> PresetReport

Tests a single preset and returns a detailed report.

Parameters:

• preset: Either a ModelPreset object or the preset name

Returns:

• PresetReport: A report containing test results including:
  • preset_name: Name of the tested preset
  • preset_data: The preset configuration
  • test_input: Test messages used
  • test_output: Model response (if successful)
  • token_prompt: Token count of input
  • token_completion: Token count of output
  • status: Whether the test succeeded
  • message: Error message (if failed)
  • time_used: Time taken for the test

Example:

report = await manager.test_single_preset("gpt-4-preset")
if report.status:
    print(f"✓ Test passed in {report.time_used:.2f}s")
else:
    print(f"✗ Test failed: {report.message}")

async test_presets() -> AsyncGenerator[PresetReport, None]

Tests all registered presets sequentially and yields reports.

Returns:

• AsyncGenerator[PresetReport, None]: An async generator yielding test reports

Example:

async for report in manager.test_presets():
    status = "✓" if report.status else "✗"
    print(f"{status} {report.preset_name}: {report.message or 'OK'}")

Recommended Usage Pattern

from amrita_core.preset import PresetManager
from amrita_core.types import ModelPreset, ModelConfig

# Initialize the manager (singleton, only needs to be called once)
manager = PresetManager()

# Add multiple presets
manager.add_preset(
    ModelPreset(
        model="gpt-3.5-turbo",
        name="fast",
        api_key="sk-xxx",
        config=ModelConfig(stream=True),
    )
)

manager.add_preset(
    ModelPreset(
        model="gpt-4", name="smart", api_key="sk-xxx", config=ModelConfig(stream=False)
    )
)

# Set a default preset (optional, but recommended)
manager.set_default_preset("fast")

# Use presets in your application
# If you don't specify a preset, get_default_preset() will auto-fallback
preset = manager.get_default_preset()  # Returns "fast" preset

Key Benefits

1. Centralized Management: All presets are stored and managed in one place
2. Singleton Pattern: Ensures consistent preset state across your application
3. Automatic Fallback: If no preset is selected, a default is automatically chosen
4. Validation: Prevents duplicate preset names and validates configurations
5. Testing: Built-in testing capability to verify preset functionality
6. Type Safety: Full type hints for better IDE support and error prevention

See Also

• ModelPreset - The underlying preset configuration class
• AmritaConfig - Overall Amrita configuration
• AgentRuntime - Using presets with the agent runtime