airtrain 0.1.4

A platform for building and deploying AI agents with structured skills

Airtrain

A powerful platform for building and deploying AI agents with structured skills and capabilities.

Features

• Structured Skills: Build modular AI skills with defined input/output schemas
• Multiple LLM Integrations: Built-in support for OpenAI and Anthropic models
• Structured Outputs: Parse LLM responses into structured Pydantic models
• Credential Management: Secure handling of API keys and credentials
• Type Safety: Full type hints and Pydantic model support
• Image Support: Handle image inputs for multimodal models
• Error Handling: Robust error handling and logging

Installation

pip install airtrain

Quick Start

1. Basic OpenAI Chat

from airtrain.integrations.openai.skills import OpenAIChatSkill, OpenAIInput

# Initialize the skill
skill = OpenAIChatSkill()

# Create input
input_data = OpenAIInput(
    user_input="Explain quantum computing in simple terms.",
    system_prompt="You are a helpful teacher.",
    max_tokens=500,
    temperature=0.7
)

# Get response
result = skill.process(input_data)
print(result.response)
print(f"Tokens Used: {result.usage['total_tokens']}")

2. Anthropic Claude Integration

from airtrain.integrations.anthropic.skills import AnthropicChatSkill, AnthropicInput

# Initialize the skill
skill = AnthropicChatSkill()

# Create input
input_data = AnthropicInput(
    user_input="Explain the theory of relativity.",
    system_prompt="You are a physics expert.",
    model="claude-3-opus-20240229",
    temperature=0.3
)

# Get response
result = skill.process(input_data)
print(result.response)
print(f"Usage: {result.usage}")

3. Structured Output with OpenAI

from pydantic import BaseModel
from typing import List
from airtrain.integrations.openai.skills import OpenAIParserSkill, OpenAIParserInput

# Define your response model
class PersonInfo(BaseModel):
    name: str
    age: int
    occupation: str
    skills: List[str]

# Initialize the parser skill
parser_skill = OpenAIParserSkill()

# Create input with response model
input_data = OpenAIParserInput(
    user_input="Tell me about John Doe, a 30-year-old software engineer who specializes in Python and AI",
    system_prompt="Extract structured information about the person.",
    response_model=PersonInfo
)

# Get structured response
result = parser_skill.process(input_data)
person_info = result.parsed_response
print(f"Name: {person_info.name}")
print(f"Skills: {', '.join(person_info.skills)}")

Error Handling

All skills include built-in error handling:

from airtrain.core.skills import ProcessingError

try:
    result = skill.process(input_data)
except ProcessingError as e:
    print(f"Processing failed: {e}")

Advanced Features

• Image Analysis Support
• Function Calling
• Custom Validators
• Async Processing
• Token Usage Tracking

For more examples and detailed documentation, visit our documentation.

Documentation

For detailed documentation, visit our documentation site.

Telemetry

Airtrain collects telemetry data to help improve the library. The data collected includes:

• Agent run information (model used, task description, environment settings)
• Agent steps and actions (full action details and reasoning)
• Performance metrics (token usage, execution time, CPU and memory usage)
• System information (OS, Python version, machine details)
• Error information (complete stack traces and context)
• Model usage details (prompts, responses, parameters)

The telemetry helps us identify usage patterns, troubleshoot issues, and improve the library based on real-world usage. The user ID is stored at ~/.cache/airtrain/telemetry_user_id.

Disabling Telemetry

Telemetry is enabled by default, but you can disable it if needed:

1. Set an environment variable:

export AIRTRAIN_TELEMETRY_ENABLED=false

2. In your Python code:

import os
os.environ["AIRTRAIN_TELEMETRY_ENABLED"] = "false"

Viewing Telemetry Debug Information

To see what telemetry data is being sent:

os.environ["AIRTRAIN_LOGGING_LEVEL"] = "debug"

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

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