zen-guard-stream

Streaming Moderation

A 4B dense transformer optimized for low-latency streaming content moderation. Evaluates text in real time as it is generated or received, enabling token-by-token safety checks in production pipelines.

Specifications

Capabilities

• Real-time streaming content moderation
• Token-by-token safety evaluation
• Low-latency classification (< 10ms per chunk)
• Multi-category harm detection
• Configurable sensitivity thresholds
• Pipeline integration for LLM output filtering

Usage

API

from hanzoai import Hanzo

client = Hanzo(api_key="hk-your-api-key")

response = client.moderations.create(
    model="zen-guard-stream",
    input="Text to moderate for safety.",
)

for result in response.results:
    print(f"Flagged: {result.flagged}")
    for category, score in result.category_scores.items():
        if score > 0.5:
            print(f"  {category}: {score:.3f}")

See Also

• zen3-guard -- 4B safety classifier
• zen-guard-gen -- 8B safe generation model
• zen3-nano -- 4B general-purpose edge model