promptfile

LLM prompts as versioned YAML files — git-trackable, renderable, and diffable.
No servers. No SaaS. No database. Works with OpenAI, Anthropic, LiteLLM, and any string-based API.

---

Why promptfile?

Most teams manage prompts as hardcoded strings, littered through source files. When a prompt changes, there's no diff, no version history, no review process — just a mystery regression.

promptfile treats your prompts like code: plain YAML files, committed to git, reviewed in PRs, and rendered at runtime.

Problem	Without promptfile	With promptfile
Prompt lives in source code	Hard to find & change	Separate .prompt.yaml file
No version history	Git doesn't know what changed	Full git diff, blame, history
Variables scattered everywhere	f"Summarise {article}" strings	{{article}} in YAML
Can't review prompt changes	Buried in code diffs	Clean YAML diff in PR
Works only with one LLM SDK	Tied to one client	Plain dict output, any SDK

---

Install

# Core install
pip install promptfile

# With YAML parsing (recommended — you almost always need this)
pip install "promptfile[yaml]"

# Everything
pip install "promptfile[all]"

---

Quick Start

1. Write a prompt file (prompts/summarise.prompt.yaml):

name: summarise
version: "1.0.0"
description: Summarise an article in a given tone.

system: You are a professional editor. Summarise text clearly and accurately.
user: "Summarise the following article in a {{tone}} tone:\n\n{{article}}"

2. Load and render it in Python:

from promptfile import load

prompt = load("prompts/summarise.prompt.yaml")

# Render variables
rendered = prompt.render(
    tone="concise",
    article="Today's top story is about AI...",
)

# Use with any LLM client
messages = rendered.to_messages()
# → [{'role': 'system', 'content': '...'}, {'role': 'user', 'content': '...'}]

3. Pass to your LLM:

# OpenAI
import openai
client = openai.OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=rendered.to_messages(),
)

# Anthropic
import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-opus-4-6",
    max_tokens=1024,
    system=rendered.messages[0].content,
    messages=rendered.to_messages()[1:],
)

# LiteLLM
import litellm
response = litellm.completion(
    model="gpt-4o",
    messages=rendered.to_messages(),
)

That's it. Your prompt is now a versioned, reviewable, git-trackable artifact.

---

YAML Formats

Shorthand (quick authoring)

name: my-prompt
version: "1.0.0"
description: Optional human-readable description.

system: You are a helpful assistant.
user: "Answer this question: {{question}}"

Full messages list (multi-turn / precise control)

name: customer-support
version: "2.1.0"
description: Customer support agent for SaaS product.

messages:
  - role: system
    content: |
      You are a friendly customer support agent for Acme Corp.
      Always be empathetic and solution-focused.
  - role: user
    content: "{{customer_message}}"

With metadata (model hints, tags, defaults)

name: code-review
version: "1.3.0"
description: Reviews a code diff and suggests improvements.

system: You are an expert software engineer.
user: "Review this {{language}} code:\n\n{{code}}"

# Metadata: anything extra is stored in prompt.metadata
model: gpt-4o
temperature: 0.2
tags:
  - engineering
  - code-quality

defaults:
  language: Python

---

All Features

load(path) — Load a single prompt file

from promptfile import load

prompt = load("prompts/summarise.prompt.yaml")

print(prompt.name)        # "summarise"
print(prompt.version)     # "1.0.0"
print(prompt.variables)   # {"tone", "article"}
print(prompt.metadata)    # {"model": "gpt-4o", ...}

load_dir(directory) — Load all prompts in a folder

from promptfile import load_dir

prompts = load_dir("prompts/", recursive=True)
# → {"summarise": Prompt(...), "code-review": Prompt(...), ...}

summarise = prompts["summarise"]

render(prompt, **kwargs) — Fill template variables

from promptfile import load, render

prompt = load("prompts/summarise.prompt.yaml")

rendered = render(prompt, tone="formal", article="The article text...")

# Or call directly on the prompt object:
rendered = prompt.render(tone="formal", article="The article text...")

Missing variables raise a clear KeyError:

KeyError: Prompt 'summarise' requires variables: ['article', 'tone']
Provided: []

registry — Central prompt store

from promptfile import registry

# Load at startup
registry.load_dir("prompts/")

# Access anywhere in your app
summarise = registry.get("summarise")
rendered = summarise.render(tone="concise", article="...")

# List all registered prompts
print(registry.list())
# → ["code-review", "customer-support", "summarise"]

diff_prompts(old, new) — Human-readable prompt diff

from promptfile import load, diff_prompts

old = load("prompts/v1/summarise.prompt.yaml")
new = load("prompts/v2/summarise.prompt.yaml")

print(diff_prompts(old, new, color=True))

--- summarise v1.0.0
+++ summarise v2.0.0
@@ -3,4 +3,4 @@
   - role: user
-    Summarise the following article in a {{tone}} tone:
+    Summarise the following article in a {{tone}} tone. Be {{length}}.

validate(prompt) — Catch authoring mistakes

from promptfile import load
from promptfile._validate import validate

prompt = load("prompts/my-prompt.yaml")
result = validate(prompt)

if not result.valid:
    for error in result.errors:
        print(f"ERROR: {error}")
for warning in result.warnings:
    print(f"WARN:  {warning}")

---

CLI

# Show a prompt
promptfile show prompts/summarise.prompt.yaml

# Validate all prompts in a directory
promptfile validate prompts/ --recursive

# Diff two versions
promptfile diff prompts/v1/summarise.yaml prompts/v2/summarise.yaml

# List all prompts
promptfile list prompts/ --recursive

# Render a prompt with variables (for manual testing)
promptfile render prompts/summarise.prompt.yaml \
    --var tone=concise \
    --var article="The quick brown fox."

---

Git Integration

Because prompt files are plain YAML, they integrate naturally with git:

# See what changed in your prompts
git diff HEAD~1 prompts/

# Blame a specific line
git blame prompts/summarise.prompt.yaml

# Review prompt changes in a PR — it's just YAML

Recommended project layout:

my-project/
├── prompts/
│   ├── summarise.prompt.yaml
│   ├── code-review.prompt.yaml
│   ├── customer-support.prompt.yaml
│   └── v2/
│       └── summarise.prompt.yaml   ← in-progress update
├── src/
│   └── ...
└── tests/
    └── test_prompts.py

---

Testing Prompts

Use promptfile with genassert for semantic prompt testing:

import pytest
from promptfile import load
from genassert import assert_intent, assert_no_hallucination

PRODUCT_FACTS = [
    "The price is $49/month",
    "There is a 14-day free trial",
]

def test_support_prompt():
    prompt = load("prompts/customer-support.prompt.yaml")
    rendered = prompt.render(customer_message="How much does it cost?")

    # Call your LLM here...
    response = my_llm(rendered.to_messages())

    assert_intent(response, "pricing information")
    assert_no_hallucination(response, PRODUCT_FACTS)

---

CI Integration

# .github/workflows/prompts.yml
name: Validate Prompts

on: [push, pull_request]

jobs:
  validate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.11"
      - run: pip install "promptfile[yaml]"
      - run: promptfile validate prompts/ --recursive

---

Framework Compatibility

promptfile outputs plain Python dicts — it works with every LLM client:

messages = rendered.to_messages()
# → [{"role": "system", "content": "..."}, {"role": "user", "content": "..."}]

# OpenAI
openai_client.chat.completions.create(model="gpt-4o", messages=messages)

# Anthropic
anthropic_client.messages.create(model="claude-opus-4-6", messages=messages[1:], system=messages[0]["content"])

# LiteLLM
litellm.completion(model="gpt-4o", messages=messages)

# LangChain
from langchain_core.messages import SystemMessage, HumanMessage
lc_messages = [SystemMessage(m["content"]) if m["role"] == "system" else HumanMessage(m["content"]) for m in messages]

---

Configuration

No configuration files needed. Optional environment variables:

Variable	Default	Description
PROMPTFILE_DIR	—	Default prompts directory for registry.load_dir()

---

License

MIT © promptfile contributors

---

Related Projects

• genassert — pytest-native semantic testing for LLM apps
• PyYAML — YAML parsing
• LiteLLM — unified LLM client

---

promptfile is the missing piece between your prompts and your version control.
Stop managing prompts as strings. Start managing them as files.