kiarina-llm 1.5.0

LLM utilities and context management for kiarina namespace

kiarina-llm

A Python library for LLM utilities and context management with type safety and configuration management.

Features

• RunContext Management: Structured context information for LLM pipeline processing
• Type Safety: Full type hints and Pydantic validation
• Configuration Management: Use pydantic-settings-manager for flexible configuration
• Filesystem Safe Names: Validated names for cross-platform compatibility
• ID Validation: Structured ID types with pattern validation

Installation

pip install kiarina-llm

Quick Start

Basic RunContext Usage

from kiarina.llm.run_context import create_run_context

# Create a run context with default settings
context = create_run_context(
    tenant_id="tenant-123",
    user_id="user-456",
    agent_id="my-agent",
    time_zone="Asia/Tokyo",
    language="ja"
)

print(f"User: {context.user_id}")
print(f"Agent: {context.agent_id}")
print(f"Time Zone: {context.time_zone}")
print(f"Language: {context.language}")

Configuration Management

from kiarina.llm.run_context import settings_manager

# Configure default values
settings_manager.user_config = {
    "app_author": "MyCompany",
    "app_name": "MyAIApp",
    "tenant_id": "default-tenant",
    "user_id": "default-user",
    "time_zone": "America/New_York",
    "language": "en"
}

# Create context with configured defaults
context = create_run_context(
    agent_id="specialized-agent"  # Override only specific values
)

Environment Variable Configuration

Configure defaults using environment variables:

export KIARINA_LLM_RUN_CONTEXT_APP_AUTHOR="MyCompany"
export KIARINA_LLM_RUN_CONTEXT_APP_NAME="MyAIApp"
export KIARINA_LLM_RUN_CONTEXT_TENANT_ID="prod-tenant"
export KIARINA_LLM_RUN_CONTEXT_TIME_ZONE="Asia/Tokyo"
export KIARINA_LLM_RUN_CONTEXT_LANGUAGE="ja"

RunContext Fields

The RunContext model includes the following fields:

Field	Type	Description	Example
app_author	FSName	Application author (filesystem safe)	"MyCompany"
app_name	FSName	Application name (filesystem safe)	"MyAIApp"
tenant_id	IDStr	Tenant identifier	"tenant-123"
user_id	IDStr	User identifier	"user-456"
agent_id	IDStr	Agent identifier	"my-agent"
runner_id	IDStr	Runner identifier	"linux" (auto-detected)
time_zone	str	IANA time zone	"Asia/Tokyo"
language	str	ISO 639-1 language code	"ja"
metadata	dict[str, Any]	Additional metadata	{"version": "1.0"}

Type Validation

FSName (Filesystem Safe Name)

The FSName type ensures names are safe for use across different filesystems:

from kiarina.llm.run_context import create_run_context

# Valid names
context = create_run_context(
    app_author="My Company",      # Spaces allowed
    app_name="My-App_v1.0"       # Hyphens, underscores, dots allowed
)

# Invalid names (will raise ValidationError)
try:
    create_run_context(app_author="My App.")  # Ends with dot
except ValueError as e:
    print(f"Validation error: {e}")

try:
    create_run_context(app_author=".hidden")  # Starts with dot
except ValueError as e:
    print(f"Validation error: {e}")

try:
    create_run_context(app_author="CON")  # Windows reserved name
except ValueError as e:
    print(f"Validation error: {e}")

IDStr (ID String)

The IDStr type validates identifiers:

# Valid IDs
context = create_run_context(
    tenant_id="tenant-123",
    user_id="user.456",
    agent_id="agent_v1.0"
)

# Invalid IDs (will raise ValidationError)
try:
    create_run_context(tenant_id="")  # Empty string
except ValueError as e:
    print(f"Validation error: {e}")

try:
    create_run_context(user_id="user@domain")  # Invalid character
except ValueError as e:
    print(f"Validation error: {e}")

Advanced Usage

Custom Metadata

context = create_run_context(
    tenant_id="tenant-123",
    user_id="user-456",
    metadata={
        "session_id": "session-789",
        "request_id": "req-abc123",
        "version": "1.0.0",
        "features": ["feature-a", "feature-b"]
    }
)

print(f"Session: {context.metadata['session_id']}")
print(f"Features: {context.metadata['features']}")

Integration with PlatformDirs

The app_author and app_name fields are designed to work with libraries like platformdirs:

from platformdirs import user_data_dir
from kiarina.llm.run_context import create_run_context

context = create_run_context(
    app_author="MyCompany",
    app_name="MyAIApp"
)

# Use with platformdirs
data_dir = user_data_dir(
    appname=context.app_name,
    appauthor=context.app_author
)
print(f"Data directory: {data_dir}")

Configuration Reference

Setting	Environment Variable	Default	Description
app_author	KIARINA_LLM_RUN_CONTEXT_APP_AUTHOR	"kiarina"	Default application author
app_name	KIARINA_LLM_RUN_CONTEXT_APP_NAME	"myaikit"	Default application name
tenant_id	KIARINA_LLM_RUN_CONTEXT_TENANT_ID	""	Default tenant ID
user_id	KIARINA_LLM_RUN_CONTEXT_USER_ID	""	Default user ID
agent_id	KIARINA_LLM_RUN_CONTEXT_AGENT_ID	""	Default agent ID
runner_id	KIARINA_LLM_RUN_CONTEXT_RUNNER_ID	platform.system().lower()	Default runner ID
time_zone	KIARINA_LLM_RUN_CONTEXT_TIME_ZONE	"UTC"	Default time zone
language	KIARINA_LLM_RUN_CONTEXT_LANGUAGE	"en"	Default language

Development

Prerequisites

• Python 3.12+

Setup

# Clone the repository
git clone https://github.com/kiarina/kiarina-python.git
cd kiarina-python

# Setup development environment (installs tools, syncs dependencies, downloads test data)
mise run setup

Running Tests

# Run format, lint, type checks and tests
mise run package kiarina-llm

# Coverage report
mise run package:test kiarina-llm --coverage

# Run specific tests
uv run --group test pytest packages/kiarina-llm/tests/run_context/

Dependencies

• pydantic - Data validation using Python type hints
• pydantic-settings - Settings management
• pydantic-settings-manager - Advanced settings management

Roadmap

This package is in active development. Planned features include:

• Chat Model Management: Unified interface for different LLM providers
• Agent Framework: Tools for building LLM agents
• Pipeline Management: Workflow management for LLM processing
• Memory Management: Context and conversation memory handling
• Tool Integration: Framework for LLM tool calling

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributing

This is a personal project, but contributions are welcome! Please feel free to submit issues or pull requests.

Related Projects

• kiarina-python - The main monorepo containing this package
• pydantic-settings-manager - Configuration management library used by this package