Core Concepts

PromptKit is built around structured prompt engineering - treating prompts as first-class artifacts with schemas, validation, and reusability.

Key Components

Prompts

A Prompt contains:

Template: Jinja2 template with variables
Schema: Type definitions for input validation
Metadata: Name and description

from promptkit import Prompt, load_prompt

# Load from YAML
prompt = load_prompt("greeting.yaml")

# Or create programmatically
prompt = Prompt(
    name="greeting",
    description="Personal greeting",
    template="Hello {{ name }}!",
    input_schema={"name": "str"}
)

Engines

Engines provide a unified interface to LLM providers:

from promptkit import OpenAIEngine
from promptkit.engines import OllamaEngine

# Cloud
engine = OpenAIEngine(api_key="sk-...", model="gpt-4o-mini")

# Local  
engine = OllamaEngine(model="llama2")

Runner

The runner orchestrates prompt execution:

from promptkit import run_prompt

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

Workflow

Basic Execution

from promptkit import load_prompt, run_prompt, OpenAIEngine

prompt = load_prompt("greeting.yaml")
engine = OpenAIEngine(api_key="sk-...")
response = run_prompt(prompt, {"name": "Alice"}, engine)

Template Preview

prompt = load_prompt("greeting.yaml")
rendered = prompt.render({"name": "Alice"})
print(rendered)

Validation

Inputs are validated against schemas before rendering:

input_schema:
  name: str           # Required string
  age: int            # Required integer
  email: "str | None" # Optional string

YAML Format

name: my_prompt
description: What this prompt does
template: |
  Hello {{ name }}!
  {% if context %}
  Context: {{ context }}
  {% endif %}
input_schema:
  name: str
  context: "str | None"