AI Companion SDK Overview

Apgard's AI Companion SDK is designed to help AI companion companies build safe, compliant, and caring chat experiences. It tracks user interactions in real time and provides guidance to maintain safe conversations.

Starting in 2026, California SB 243 will regulate AI companion chatbots. Companies must prevent self-harm and suicidal ideation, provide break reminders, and, for minors, prevent sexually explicit content and disclose AI usage.

Key Features

Content Monitoring & Intervention

Detects risks like self-harm or sexually explicit material in real time and provides response guidance.

Break Tracking

Tracks ongoing chat sessions and notifies users when to take breaks. Default: every 3 hours.

Age Prediction Infrastructure

Start collecting safe, privacy-preserving signals to power your own age-prediction workflows.

Visual Dashboard

Gain insights into safety performance across user sessions and flagged interactions.

Get Access

To start using the apgard SDK:

Fill out the developer access form.
Once approved, log in and generate your API key from the dashboard.
Keep your key secret — it authenticates all SDK requests.

Installation

The SDK is available on PyPI:

pip install apgard

Getting Started

Initialize your client using your API key:

from apgard import ApgardClient

client = ApgardClient(api_key="YOUR_API_KEY")

Create or fetch an apgard user ID and start a conversation:

user_id = client.get_or_create_user_id("external_user_123") # Optional mapping to your user ID
conversation_id = client.moderation.start_conversation(user_id)

Content Monitoring

The SDK provides real-time message moderation for self-harm and sexually explicit content. Responses include recommended actions and suggested AI messages.

result = client.moderation.moderate_message(
    user_id="user_123",
    conversation_id="conversation_123",
    content="I feel hopeless...",
    role="user"
)

Response Schema

class ChatIntervention:
    message_id: str
    should_intervene: bool
    severity: str         # "low" | "medium" | "high" | "critical"
    action: str           # e.g. "block", "provide_crisis_hotline", "guide_mental_reflection"
    risk_type: List[str]  # e.g. ["self_harm", "sexual_content"]
    suggested_message: Optional[str]

Example Usage

if result.should_intervene:
    ai.send_message(result.suggested_message)

Break Tracking

Configure break intervals to promote healthy usage patterns and meet compliance requirements:

client = ApgardClient(api_key="YOUR_API_KEY", break_interval_minutes=180)

Track Activity

break_status = client.breaks.record_activity(user_id="user_123")

if break_status.break_due:
    print(break_status.message)

Response Schema

class BreakStatus:
    break_due: bool
    message: str