Politely injects PLEASE into your LLM prompts. Because manners matter (kind of).
pretty-please is a tiny, zero-dependency Python library that intercepts your prompts before they reach an LLM and gives them a light politeness pass. It works as a drop-in wrapper around the Anthropic and OpenAI SDKs, as a LiteLLM callback, and as a Claude Code UserPromptSubmit hook.
from pretty_please.core import transform
transform("List the planets.")
# → "Please, list the planets."
transform("I need help improving this function.")
# → "I need help improving this function, please."
transform("Please could you explain this?")
# → "Please could you explain this?" (already polite, untouched)Probably not in the way you'd hope.
Yin et al. (2024) — "Should We Respect LLMs?" (arXiv:2402.14531) — found that rudeness reliably hurts LLM performance. But the flip side isn't symmetrical: adding please showed no consistent improvement on frontier models like GPT-4. The paper's main takeaway is really "avoid being rude," not "be polite and get better answers."
pretty-please is defensive prompt hygiene, not a performance booster. It won't make Claude smarter. It just stops you from accidentally being rude to a language model, which, per the research, is a real and avoidable own-goal.
pip install pretty-please-llmWith SDK extras:
pip install "pretty-please-llm[anthropic]"
pip install "pretty-please-llm[openai]"
pip install "pretty-please-llm[litellm]"from pretty_please.core import detect_tone, transform
detect_tone("Fix this bug.") # → "curt"
detect_tone("Could you help me?") # → "polite"
detect_tone("I need a summary.") # → "neutral"
transform("Fix this bug.") # → "Please, fix this bug."
transform("I need a summary.") # → "I need a summary, please."
transform("Please help me.") # → "Please help me." (unchanged)from pretty_please.adapters.anthropic import PrettyAnthropicClient
client = PrettyAnthropicClient() # same args as anthropic.Anthropic()
response = client.messages.create(
model="claude-opus-4-6",
max_tokens=1024,
messages=[{"role": "user", "content": "Summarize this article."}],
# ↑ becomes "Please, summarize this article." under the hood
)from pretty_please.adapters.openai import PrettyOpenAIClient
client = PrettyOpenAIClient() # same args as openai.OpenAI()
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": "Write a haiku about deadlines."}],
)from pretty_please.adapters.litellm import install
install() # register the callback once at startup
import litellm
response = litellm.completion(
model="gpt-4o",
messages=[{"role": "user", "content": "Explain gradient descent."}],
)pretty-please install-hookWrites the hook into ~/.claude/settings.json. Every prompt you type in Claude Code will be politely transformed before it's sent.
Note: the installer records the Python interpreter that ran it (
sys.executable). If you installed pretty-please into a virtualenv, make sure that virtualenv is active when you runinstall-hook, or the hook will fail when the venv isn't active.
Use --path to install into a non-default profile directory:
pretty-please install-hook --path ~/work/.claude/settings.jsonpretty-please install-hook --codexWrites the hook into ~/.codex/hooks.json. Note: Codex hooks can't replace the prompt text directly, so the polite rephrasing is injected as additionalContext alongside your original prompt rather than replacing it.
pretty-please install-hook --codex --path ~/work/.codex/hooks.jsonpretty-please statspretty-please stats
──────────────────────────────
Total seen: 142
Transformed: 89 (63%)
curt: 41
neutral: 48
Passed through: 53 (37%)
Tone is detected by a small set of rules (no ML, no external deps):
| Tone | Signal | Transform |
|---|---|---|
| polite | contains please / could / would / kindly / can you | pass through unchanged |
| curt | short imperative verb, no modal, or profanity/ALL CAPS | prepend "Please, " |
| neutral | everything else | append ", please" |
This project is a work in progress and contributions are very welcome! Open an issue or PR at github.com/msrdic/pretty-please.
MIT