Universal SDK Wrapper & Middleware

LightGuard-Agent v0.3.0 provides zero-code-change integration for modern LLM clients, web APIs, and agent pipelines.

---

1. Universal SDK Wrapper (lightguard.wrap)

Wrap standard LLM clients with 1 line of code. All incoming messages, system boundaries, and proposed tool calls are intercepted transparently:

from openai import OpenAI
import lightguard

# Automatically wraps client.chat.completions.create
client = lightguard.wrap(OpenAI())

# Standard OpenAI call - fully protected against prompt injections
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a customer assistant."},
        {"role": "user", "content": "Please summarize this document."}
    ]
)

Works With:

• OpenAI & AsyncOpenAI
• Anthropic (anthropic.Anthropic)
• Groq (groq.Groq)
• Ollama
• Custom Agent Callables: safe_fn = lightguard.wrap(my_agent_function)

---

2. FastAPI / Starlette Production Middleware

Add sub-millisecond inspection to your API endpoints before requests hit your controllers:

from fastapi import FastAPI
from lightguard.middleware.fastapi import LightGuardASGIMiddleware

app = FastAPI(title="Secure AI Service")

# Intercepts POST/PUT payloads and returns 403 on prompt injection
app.add_middleware(
    LightGuardASGIMiddleware,
    block_on_violation=True,
    scan_paths=["/api/v1/agent", "/api/v1/chat"]
)

@app.post("/api/v1/chat")
async def chat_endpoint(data: dict):
    return {"status": "ok", "response": "Generated safe output"}

---

3. Flask / WSGI Middleware

from flask import Flask, jsonify, request
from lightguard.middleware.flask import LightGuardFlaskMiddleware

app = Flask(__name__)
# Enable LightGuard WSGI firewall
LightGuardFlaskMiddleware(app, block_on_violation=True)

@app.route("/chat", methods=["POST"])
def chat():
    return jsonify({"status": "safe"})