pydefend 1.0.0

AI security guardrails for LLM applications — scan inputs and check outputs with Claude, OpenAI, Gemini, Azure, or Ollama.

pydefend

AI security guardrails for LLM applications — scan inputs and check outputs with Claude, OpenAI, Gemini, Azure, or Ollama.

For the full project README, docs, and examples, see:

• GitHub repo: https://github.com/Adxzer/defend
• Docs: https://www.pydefend.com/docs

Install

pip install pydefend

Optional provider extras:

pip install "pydefend[openai]"      # OpenAI SDK integration
pip install "pydefend[anthropic]"   # Anthropic SDK integration
pip install "pydefend[gemini]"      # Gemini (google-genai) SDK integration
pip install "pydefend[fastapi]"     # FastAPI middleware/integration helpers
pip install "pydefend[langchain]"   # LangChain integration helpers
pip install "pydefend[all]"         # All optional integrations

Quick start

import asyncio
from pydefend import Defend

defend = Defend.default()

async def main():
    scanned = await defend.scan("User message here")
    if not scanned:
        return "Can't help with that."

    # ... call your LLM with scanned.text ...

    checked = await defend.check("Model reply here")
    return checked.text if checked else None

print(asyncio.run(main()))

Configuration

• Code-first: use Defend.default() or the builder API to choose modules.
• Config file: drop a defend.yaml into your repo and load it with Defend.from_yaml("defend.yaml").

Links

• Repository / issues: https://github.com/Adxzer/defend
• Documentation: https://www.pydefend.com/docs

License

Apache-2.0 — see LICENSE in the GitHub repo.