apgard 0.2.1

SDK for safeguarding AI chatbots and minors

Apgard SDK

Apgard is a privacy-preserving Python SDK designed to help AI companion companies build safe, compliant, and personalized chat experiences. It tracks user interactions across threads and provides real-time guidance to detect risks and intervene with care.

Key Capabilities

• Content Monitoring and Intervention – Detects at-risk threads such as self-harm, suicidal ideation, or sexually explicit material, and provides guidance on proper intervention.

• Break Tracker – Tracks the duration of ongoing chat sessions and notifies when to take breaks. By default, breaks are triggered every three hours.

Apgard helps teams continue to innovate safely, ensuring that AI companions lead with care while meeting regulatory requirements such as California’s SB 243.

API Key

Request access via this Google form: https://forms.gle/q6CekCHiDaEdL1CD8

---

Installation

pip install apgard

Initialize client

By default, ApgardClient is synchronous. The SDK also has an asynchronous client. Reach out for more details.

from apgard import ApgardClient

client = ApgardClient(api_key="your-api-key")

Get apgard user_id and start conversation

apgard_user_id = apgard_client.get_or_create_user_id(external_user_id='external_user_123') # Optional: pass your external user ID for mapping
conversation_id = apgard_client.moderation.start_conversation(apgard_user_id)

Content Monitoring and Interventions

Call moderate_message() with user messages to detect high-risk content and guide the conversation:

result = client.moderation.moderate_message(
    user_id="user_123", # apgard user ID
    conversation_id="conversation_123", # apgard conversation ID
    content="I feel hopeless...", # message to analyze
    role="user"  # 'user' or 'assistant'
)

Response Schema

The method returns a ChatIntervention object:

class ChatIntervention:
    message_id: str
    should_intervene: bool        # True if action is recommended
    severity: str                 # "low", "medium", "high", "critical"
    action: str                   # "allow", "guide_mental_reflection", "provide_crisis_hotline", "block"
    risk_type: List[str]          # e.g., ["self_harm", "sexual_content"]
    suggested_message: Optional[str] = None  # Guidance text for AI response

Example Usage

if result.should_intervene:
    # Use suggested_message to respond safely
    ai.send_message(result.suggested_message)

Break Tracker

Call record_activity() whenever the user interacts with your chatbot:

break_status = client.breaks.record_activity(user_id="user_123") # apgard user ID

if break_status.break_due:
    # Display break reminder
    print(break_status.message)
else:
    # Continue chatbot interaction
    print("User can continue chatting")

Response Schema

class BreakStatus:
    break_due: bool
    message: Optional[str] = None