Creating and Versioning Prompts

Use push_prompt to create new prompts or update existing ones. Patronus automatically handles versioning and duplicate detection.

Creating prompts

Basic creation

import patronus
from patronus.prompts import Prompt, push_prompt
 
patronus.init()
 
prompt = Prompt(
    name="dev/bug-fix/python-error",
    body="Fix this Python code error: {error_message}. Code: ```python\n{code_snippet}\n```",
    description="Template for Python debugging assistance"
)
 
loaded_prompt = push_prompt(prompt)

With metadata

from patronus.prompts import Prompt, push_prompt
 
prompt = Prompt(
    name="support/troubleshooting/login-issues",
    body="You are a support specialist for {product_name}. Solve this issue: {issue_description}",
    description="Support prompt for login issues",
    metadata={
        "temperature": 0.7,
        "max_tokens": 500,
        "tone": "helpful"
    }
)
 
loaded_prompt = push_prompt(prompt)

Async creation

from patronus.prompts import apush_prompt
 
loaded_prompt = await apush_prompt(prompt)

How versioning works

Patronus automatically manages prompt versions:

Each time you push a prompt, Patronus checks if the content is identical to an existing revision
If the content matches an existing revision, that revision is returned (no duplicate created)
If the content is different, a new revision is created with an incremented revision number
Revisions are immutable once created

# First push creates revision 1
prompt_v1 = Prompt(
    name="support/greeting",
    body="Hello, how can I help you?"
)
loaded_v1 = push_prompt(prompt_v1)  # Creates revision 1
 
# Pushing identical content returns revision 1
loaded_again = push_prompt(prompt_v1)  # Returns existing revision 1
 
# Pushing different content creates revision 2
prompt_v2 = Prompt(
    name="support/greeting",
    body="Hello! How may I assist you today?"
)
loaded_v2 = push_prompt(prompt_v2)  # Creates revision 2

Multi-line prompts

Use textwrap.dedent for readable multi-line prompts:

import textwrap
from patronus.prompts import Prompt, push_prompt
 
prompt = Prompt(
    name="content/blog/technical-writer",
    body=textwrap.dedent("""
        You are a technical writer for {company_name}.
 
        Topic: {topic}
        Audience: {audience_level}
        Tone: {tone}
 
        Write a {word_count}-word blog post that:
        1. Introduces the topic clearly
        2. Provides practical examples
        3. Concludes with actionable takeaways
 
        Use clear, concise language appropriate for {audience_level} readers.
        """),
    description="Technical blog post writer"
)
 
loaded_prompt = push_prompt(prompt)

Naming conventions

Use a descriptive, hierarchical naming structure:

[domain]/[use-case]/[component]/[prompt-type]

Where [prompt-type] indicates the role in an LLM conversation:

system - Sets overall behavior and context
instruction - Provides specific task instructions
user - User message template
assistant - Assistant response template
few-shot - Contains example input/output pairs

Examples

# System prompt for support chat
"support/troubleshooting/diagnostic-questions/system"
 
# User message template
"support/chat/user-query/user"
 
# Few-shot examples for code generation
"dev/code-generation/python-function/few-shot"
 
# Instruction prompt for content writing
"marketing/email-campaigns/follow-up-template/instruction"

Consistent prefixes

Use consistent prefixes for related prompts:

# Onboarding flow prompts
"onboarding/chat/welcome/system"
"onboarding/chat/questions/user"
"onboarding/chat/intro/assistant"
 
# Support classifier prompts
"support/classifier/system"
"support/classifier/categories/instruction"

This makes it easier to filter and manage prompt families as your library grows.

Creating and Versioning Prompts

On this page