pixie-prompts

Code-first, type-checked prompt management. Manage prompt locally in your codebase, with Jinja rendering, variable type-hint and validations.

Demo

Try it live

Setup

In your project folder, install pixie-prompts[server] Python package:

pip install pixie-prompts[server]

Note: you can install pixie-prompts without the server extras for your production build.

Start the local dev server and open the web UI by running:

pp

Note: The web-browser would automatically open http://localhost:8000. You can also access the web UI at gopixie.ai.

To test prompts, create .env file with LLM API key(s):

# .env
OPENAI_API_KEY=...
GEMINI_API_KEY=...

Register Prompt

In your code, create a new prompt using create_prompt:

# prompts.py
from pixie.prompts import create_prompt

simple_prompt = create_prompt('simple_prompt')

Your prompt would automatically appear in the web UI after your code is saved.

Manage Prompt

You can create new version(s) of a prompt in the web UI.

Once saved from web UI, it will be assigned a new version id, and the content would be saved in your codebase at /.pixie/prompts/<prompt_name>/<version_id>.jinja.

Note: it's recommended to only edit your prompts via the web UI to get type-hint and validation.

Define Variables

For prompt that has variable(s) in it, define a class extending pixie.prompts.Variables (which extends pydantic.BaseModel. Then use the class type when registering your prompt.

# prompts.py
from pixie.prompts import Variables, create_prompt

class Person(Variables):
    name: str
    age: int

# Create a prompt with variables
typed_prompt = create_prompt('typed_prompt', Person)

Other than using dict, you can define your variable class in anyway that's permissible in Pydantic. I.e. you can define your field as basic types such as str, int, bool, you can have a list of permissible items, you can use Union type, and you can have nested Variable field.

The web UI will parse your variable definitions and use it to decide input fields, type-hints and validations.

Use Prompt

Compile your prompt into string with the compile function on the prompt object. Pass in the Variables object (if defined) for your prompt as argument.

# demo.py

from pixie.prompts import Variables, create_prompt

simple_prompt = create_prompt('simple_prompt')

class Person(Variables):
    name: str
    age: int

# Create a prompt with variables
typed_prompt = create_prompt('typed_prompt', Person)

simple_prompt_str = simple_prompt.compile()
typed_prompt_str = typed_prompt.compile(Person(name="Jane", age=30))

Check out more examples.