Using the Prompt Management System (`genie.prompts`)¶

Genie Tooling provides a flexible prompt management system accessible via genie.prompts. This allows you to register, retrieve, and render prompt templates, separating prompt engineering concerns from your application logic.

Core Concepts¶

PromptInterface (genie.prompts): The facade interface for all prompt-related operations.
PromptRegistryPlugin: Responsible for storing and retrieving raw prompt template content (e.g., from a file system, database).
- Built-in: FileSystemPromptRegistryPlugin (alias: file_system_prompt_registry).
PromptTemplatePlugin: Responsible for rendering a raw template string with provided data.
- Built-in: BasicStringFormatTemplatePlugin (alias: basic_string_formatter), Jinja2ChatTemplatePlugin (alias: jinja2_chat_formatter).

Configuration¶

You configure the default prompt registry and template engine via FeatureSettings or explicit MiddlewareConfig settings.

Via FeatureSettings:

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

app_config = MiddlewareConfig(
    features=FeatureSettings(
        # ... other features ...
        prompt_registry="file_system_prompt_registry", # Default
        prompt_template_engine="jinja2_chat_formatter"   # Default
    ),
    # Configure the FileSystemPromptRegistryPlugin
    prompt_registry_configurations={
        "file_system_prompt_registry_v1": { # Canonical ID
            "base_path": "./my_application_prompts",
            "template_suffix": ".j2" # If using Jinja2 templates
        }
    }
    # No specific config needed for Jinja2ChatTemplatePlugin by default
)

Explicit Configuration:

app_config = MiddlewareConfig(
    default_prompt_registry_id="file_system_prompt_registry_v1",
    default_prompt_template_plugin_id="jinja2_chat_template_v1",
    prompt_registry_configurations={
        "file_system_prompt_registry_v1": {"base_path": "prompts"}
    }
)

Using `genie.prompts`¶

1. Getting Raw Template Content¶

# Assuming 'my_application_prompts/my_task.j2' exists
template_content = await genie.prompts.get_prompt_template_content(
    name="my_task", 
    # version="v1", # Optional version
    # registry_id="custom_registry" # Optional, if not using default
)
if template_content:
    print(f"Raw template: {template_content}")

2. Rendering a String Prompt¶

This is useful for prompts that result in a single string, often for completion LLMs.

from genie_tooling.prompts.types import PromptData

prompt_data: PromptData = {"topic": "AI ethics", "length": "short"}

# Uses default registry and default template engine
rendered_string = await genie.prompts.render_prompt(
    name="summarization_prompt", # e.g., 'my_application_prompts/summarization_prompt.txt'
    data=prompt_data
)
if rendered_string:
    print(f"Rendered string prompt: {rendered_string}")
    # Example: "Summarize AI ethics in a short paragraph."

3. Rendering a Chat Prompt¶

This is used for prompts that result in a list of ChatMessage dictionaries, suitable for chat-based LLMs. The Jinja2ChatTemplatePlugin is particularly useful here as Jinja2 can easily render structured JSON/YAML.

Example Jinja2 template (my_application_prompts/chat_style_prompt.j2):

[
    {"role": "system", "content": "You are a helpful {{ persona }}."},
    {"role": "user", "content": "Tell me about {{ subject }}."}
]

Python code:

from genie_tooling.prompts.types import PromptData

chat_data: PromptData = {"persona": "historian", "subject": "the Roman Empire"}

# Uses default registry and default template engine (configured to jinja2_chat_formatter)
chat_messages = await genie.prompts.render_chat_prompt(
    name="chat_style_prompt", 
    data=chat_data
)
if chat_messages:
    print(f"Rendered chat messages: {chat_messages}")
    # Output:
    # [
    #   {'role': 'system', 'content': 'You are a helpful historian.'},
    #   {'role': 'user', 'content': 'Tell me about the Roman Empire.'}
    # ]
    # response = await genie.llm.chat(chat_messages)

4. Listing Available Templates¶

available_templates = await genie.prompts.list_templates()
# Or for a specific registry:
# available_templates = await genie.prompts.list_templates(registry_id="my_other_registry")

for template_id in available_templates:
    print(f"- Name: {template_id['name']}, Version: {template_id.get('version', 'N/A')}, Desc: {template_id.get('description')}")

Creating Custom Prompt Plugins¶

PromptRegistryPlugin: Implement get_template_content and list_available_templates.
PromptTemplatePlugin: Implement render (for string output) and render_chat_messages (for List[ChatMessage] output).

Register your custom plugins via entry points in pyproject.toml or ensure they are discoverable via plugin_dev_dirs.

Using the Prompt Management System (genie.prompts)¶