promptkit-core 0.1.1

Structured Prompt Engineering for LLM Apps

⚙️ PromptKit

A production-grade library for structured LLM prompt engineering.


Explore the docs »

Report Bug · Request Feature

---

Why PromptKit?

Managing prompts for Large Language Models can quickly become messy. Hardcoding prompts as f-strings mixes logic with presentation, lacks validation, and makes reuse difficult.

PromptKit solves this by treating your prompts as structured, version-controlled assets. By defining prompts in YAML files, you get:

• Clean Separation: Prompt templates are separate from application code
• Safety & Reliability: Built-in validation ensures prompts receive correct inputs
• Reusability: Define a prompt once, use it anywhere
• Clarity: Human-readable format anyone can understand

Features

• 📝 Declarative: Define prompts in YAML with Jinja2 templating
• 🔍 Validation: Pydantic-based input validation before rendering
• 🏗️ Engine Abstraction: Supports OpenAI and Ollama
• 💰 Cost Estimation: Estimate token counts and costs before execution
• 🖥️ CLI: Render, run, lint, and inspect prompts from terminal
• 🧪 Typed: Full type hints for IDE support

Quick Start

Installation

pip install promptkit-core

Define a Prompt

Create greet.yaml:

name: greet
description: Generates a personalized greeting
template: |
  Hello {{ name }}!
  
  {% if context %}
  Context: {{ context }}
  {% endif %}
  
  How can I help you today?

input_schema:
  name: str
  context: "str | None"

Use in Python

from promptkit import load_prompt, run_prompt, OpenAIEngine

prompt = load_prompt("greet.yaml")

engine = OpenAIEngine(api_key="sk-...")

response = run_prompt(prompt, {"name": "Alice"}, engine)
print(response)

Use the CLI

export OPENAI_API_KEY="sk-..."

# Run the prompt
promptkit run greet.yaml --name Alice

# Render template without calling AI
promptkit render greet.yaml --name Alice

# Validate prompt structure
promptkit lint greet.yaml

# Get prompt information
promptkit info greet.yaml

# Estimate costs
promptkit cost greet.yaml --model gpt-4 --name Alice

Prompt Structure

Every prompt YAML file contains:

• name: Unique identifier
• description: Human-readable description
• template: Jinja2 template with variables
• input_schema: Type definitions for inputs

Input Schema Types

input_schema:
  name: str              # Required string
  age: int               # Required integer
  score: float           # Required float
  active: bool           # Required boolean
  tags: list             # Required list
  data: dict             # Required dictionary
  email: "str | None"    # Optional string

Engines

OpenAI

from promptkit import OpenAIEngine

engine = OpenAIEngine(
    api_key="sk-...",
    model="gpt-4o-mini",
    temperature=0.7,
    max_tokens=1000
)

Ollama (Local)

from promptkit.engines import OllamaEngine

engine = OllamaEngine(
    model="llama2",
    temperature=0.7
)

Documentation

See the documentation for detailed guides and API reference.

Contributing

Contributions welcome. See CONTRIBUTING.md for guidelines.

License

MIT License - see LICENSE for details.