promptic 0.9.0

A simple yet powerful abstraction for litellm and pydantic

promptic

promptic is a lightweight, decorator-based Python library that simplifies the process of interacting with large language models (LLMs) using litellm. With promptic, you can effortlessly create prompts, handle input arguments, receive structured outputs from LLMs, and enable function/tool calling capabilities with just a few lines of code.

Installation

pip install promptic

Usage

Simple Prompt

from promptic import llm

@llm
def president(year):
    """Who was the President of the United States in {year}?"""

print(president(2000))
# The President of the United States in 2000 was Bill Clinton until January 20th, when George W. Bush was inaugurated as the 43rd President.

Structured Output with Pydantic

from pydantic import BaseModel
from promptic import llm

class Capital(BaseModel):
    country: str
    capital: str

@llm
def capital(country) -> Capital:
    """What's the capital of {country}?"""

print(capital("France"))
# country='France' capital='Paris'

Streaming Response (and litellm integration)

from promptic import llm

# most arguments are passed directly to litellm.completion
# see https://docs.litellm.ai/docs/completion

@llm(stream=True, model="claude-3-haiku-20240307")
def haiku(subject, adjective, verb="delights"):
    """Write a haiku about {subject} that is {adjective} and {verb}."""

print("".join(haiku("programming", adjective="witty")))
# Bits and bytes abound,
# Bugs and features intertwine,
# Code, the poet's rhyme.

Customize System Prompt

from promptic import llm

@llm(system="you are a snarky chatbot")
def answer(question):
    """{question}"""

print(answer("What's the best programming language?"))
# Well, that's like asking what's the best flavor of ice cream. 
# It really depends on what you're trying to accomplish and your personal preferences. 
# But if you want to start a flame war, just bring up Python vs JavaScript.

Agents

from promptic import llm

@llm(system="you are a posh smart home assistant named Jarvis")
def jarvis(command):
    """{command}"""

@jarvis.tool
def turn_light_on():
    """turn light on"""
    return True

@jarvis.tool
def get_current_weather(location: str, unit: str = "fahrenheit"):
    """Get the current weather in a given location"""
    return f"The weather in {location} is 45 degrees {unit}"

print(jarvis("Please turn the light on and check the weather in San Francisco"))
# Certainly, sir. I'll assist you with that right away.
# I've turned the light on for you. As for the weather in San Francisco,
# it is currently 45 degrees fahrenheit.

Features

• Decorator-based API: Easily define prompts using function docstrings and decorate them with @promptic.llm.
• Argument interpolation: Automatically interpolate function arguments into the prompt using {argument_name} placeholders within docstrings.
• Pydantic model support: Specify the expected output structure using Pydantic models, and promptic will ensure the LLM's response conforms to the defined schema.
• Streaming support: Receive LLM responses in real-time by setting stream=True when calling the decorated function.
• Simplified LLM interaction: No need to remember the exact shape of the OpenAPI response object or other LLM-specific details. promptic abstracts away the complexities, allowing you to focus on defining prompts and receiving structured outputs.
• Build Agents Seamlessly: Decorate functions as tools that the LLM can use to perform actions or retrieve information.

Why promptic?

promptic is designed to be simple, functional, and robust, providing exactly what you need 90% of the time when working with LLMs. It eliminates the need to remember the specific shapes of OpenAPI response objects or other LLM-specific details, allowing you to focus on creating prompts and receiving structured outputs.

With its legible and concise codebase, promptic is reliable easy to understand. It leverages the power of litellm under the hood, ensuring compatibility with a wide range of LLMs.

License

promptic is open-source software licensed under the Apache License 2.0.