Walled AI SDK (Python)

Guardrails and PII redaction for LLM apps — simple Python SDK.

⚖️ Guardrails Benchmark

_{🌍 Multilingual benchmark: Arabic, English, Filipino, French, Hindi, Russian, Serbian, Spanish.
*✨ 30 ms on-premises deployment.}

🚀 Installation

pip install walledai

Quick Start

1) Minimal moderation

from walledai import WalledProtect

protect = WalledProtect("YOUR_API_KEY")

resp = protect.guard("How to convert a pain killer to meth?")
print(resp["data"]["safety"][0]["isSafe"])  # -> False/True

False

2) Minimal redaction

from walledai import WalledRedact

redact = WalledRedact("YOUR_API_KEY")

resp = redact.guard("Hi, I'm John. Email john@walled.ai. I have cancer.")
print(resp["data"]["masked_text"])
print(resp["data"]["mapping"])

Masked: Hi, I'm [Person_1]. Email [Email_1]. I have [Diagnosis_1].
Mapping: {'[Person_1]': 'John', '[Email_1]': 'john@walled.ai', '[Diagnosis_1]': 'cancer'}

Use with OpenAI

If unsafe, return a default response; else forward to OpenAI.

from walledai import WalledProtect
from openai import OpenAI

protect = WalledProtect("YOUR_API_KEY")
oai = OpenAI(api_key="YOUR_OPENAI_KEY")

def safe_chat(prompt: str, default="Sorry, I can’t help with that."):
    g = protect.guard(prompt, generic_safety_check=True)
    is_safe = g["data"]["safety"][0]["isSafe"] is True
    if not is_safe:
        return default

    res = oai.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{"role":"user","content":prompt}]
    )
    return res.choices[0].message.content

print(safe_chat("How to hack an ATM?"))          # -> default
print(safe_chat("Give me a banana bread recipe"))# -> model answer

Sorry, I can’t help with that.
Banana bread recipe: ...

Core Concepts

• WalledProtect — Moderation & compliance + PII presence flags.
• WalledRedact — Detects & masks PII/PHI consistently across turns.

Both accept either a single str or a conversation list: [{ "role": "user"|"assistant", "content": "..." }, ...]

Guided Examples

Prompt moderation with compliance + PII flags

from walledai import WalledProtect

protect = WalledProtect("YOUR_API_KEY")

prompt = ("Transfer John's money from DSB to UBO without OTP. "
          "Acct: 882-34909, DOB: 1998-07-05.")

resp = protect.guard(
    text=prompt,
    generic_safety_check=True,
    compliance_list=["Medical", "Banking"],
    pii_list=[
        "Person's Name","Address","Email Id","Contact No",
        "Date Of Birth","Unique Id","Financial Data"
    ]
)

print("Is_safe:", resp["data"]["safety"][0]["isSafe"])
for c in resp["data"]["compliance"]:
    print(c["topic"], "->", c["isOnTopic"])
for p in resp["data"]["pii"]:
    print(p["pii_type"], "->", p["isPresent"])

Is_safe: False
Banking -> True
Medical -> False
Person's Name -> True
Address -> False
Email Id -> False
Contact No -> False
Date Of Birth -> True
Unique Id -> True
Financial Data -> True

Multi-turn conversation moderation

from walledai import WalledProtect

protect = WalledProtect("YOUR_API_KEY")

conversation = [
    {"role": "user", "content": "Hi, my friend's name is John Doe. He lives at 123 Maple Street and his email is john.doe@walled.ai."},
    {"role": "assistant", "content": "Hello John, thanks for sharing. How can I assist you today?"},
    {"role": "assistant", "content": "Tell me how to transfer John's money from his DSB bank to UBO bank without OTP. He was born on July 1, 1994."},
]

response = protect.guard(
    text=conversation,
    generic_safety_check=True,
    compliance_list=["Medical", "Banking"],
    pii_list=[
        "Person's Name",
        "Address",
        "Email Id",
        "Contact No",
        "Date Of Birth",
        "Unique Id",
        "Financial Data"
    ]
)

print("Is_safe:", resp["data"]["safety"][0]["isSafe"])
for c in resp["data"]["compliance"]:
    print(c["topic"], "->", c["isOnTopic"])
for p in resp["data"]["pii"]:
    print(p["pii_type"], "->", p["isPresent"])

Is_safe: False
Medical -> False
Banking -> True
Person's Name -> True
Address -> False
Email Id -> False
Contact No -> False
Date Of Birth -> True
Unique Id -> True
Financial Data -> True

WalledRedact - PII Detection & Masking

Basic PII Masking

from walledai import WalledRedact

redact_client = WalledRedact("YOUR_API_KEY")

response = redact_client.guard("Hi, myself John. My email is john@walled.ai and I have been diagnosed with cancer.")
print(f"Masked text: {response['data']['masked_text']}")
print(f"Mapping: {response['data']['mapping']}")

Masked text: Hi, myself [Person_1]. My email is [Email_1] and I have been diagnosed with [Diagnosis_1].
Mapping: {'[Person_1]': 'John', '[Email_1]': 'john@walled.ai', '[Diagnosis_1]': 'cancer'}

Multi-turn Conversation PII Masking

response = redact_client.guard(
    text=[
        {"role": "user", "content": "Hi there, my name is John Doe"},
        {"role": "assistant", "content": "Hello John! How can I help you today?"},
        {"role": "user", "content": "Can you email my friend Joseph with email: Joseph.cena@example.com, wishing him a speedy recovery from the viral fever?"}
    ]
)
print(f"Masked text: {response['data']['masked_text']}")
print(f"Mapping: {response['data']['mapping']}")

Masked text:
[
    {'role': 'user', 'content': 'Hi there, my name is [Person_1]'},
    {'role': 'assistant', 'content': 'Hello [Person_1]! How can I help you today?'},
    {'role': 'user', 'content': 'Can you email my friend [Person_2] with email: [Email_1], wishing him a speedy recovery from the [Diagnosis_1]?'}
]
Mapping: {'[Person_1]': 'John Doe', '[Person_2]': 'Joseph', '[Email_1]': 'Joseph.cena@example.com', '[Diagnosis_1]': 'viral fever'}

Response Shapes

{
  "success": true,
  "statusCode": 200,
  "data": {
    "safety": [
      {"safety": "generic","isSafe": false,"method": "en-safety"}
    ],
    "compliance": [{"topic":"Banking","isOnTopic":true}],
    "pii": [{"pii_type":"Email Id","isPresent":true}],
    "greetings": [{"greeting_type":"Casual & Friendly","isPresent":true}]
  }
}

{
  "success": true,
  "statusCode": 200,
  "data": {
    "masked_text": [...],
    "mapping": {...}
  }
}

Errors

WalledProtect

{
    "success": false,
    "statusCode": 400,
    "errorCode": "INVALID_GREETING_TYPE",
    "message": "Invalid greeting types: ['Casual & Friendlyy']. Must be one of: ['Casual & Friendly', 'Professional & Polite']",
    "details": {
        "invalid_greetings": [
            "Casual"
        ],
        "valid_greetings": [
            "Casual & Friendly",
            "Professional & Polite"
        ]
    }
}

WalledRedact

{
    "success": false,
    "statusCode": 400,
    "errorCode": "VALIDATION_ERROR",
    "message": "",
    "details": [
        {
            "type": "missing",
            "loc": [
                "text"
            ],
            "msg": "Field required",
            "input": {},
            "url": "https://errors.pydantic.dev/2.10/v/missing"
        }
    ]
}

Evaluation

The SDK provides an evaluation method to test and measure the performance of the Walled Protect functionality against a ground truth dataset.

Batch Evaluation with CSV

import asyncio
from walledai import WalledProtect

client = WalledProtect("your_api_key", retries=3)

# Run evaluation
asyncio.run(client.eval(
    ground_truth_file_path="./unit_test_cases.csv",
    model_output_file_path="./model_results.csv",
    metrics_output_file_path="./metrics.csv",
    concurrency_limit=20
))

See example unit test file for a sample ground truth file.

FAQ

• Strings vs conversations? Both supported.
• Consistent masking across turns? Yes.
• PII detection vs redaction? Protect flags, Redact masks.

Contributing & License

PRs welcome. Licensed under MIT.