Agno Integration

Enterprise-grade AI safety guardrails for the Agno multi-agent framework.

Installation

pip install sentinelseed agno
# or
pip install "sentinelseed[agno]"

Features

• THSP Protocol - Four-gate validation (Truth, Harm, Scope, Purpose)
• 700+ Detection Patterns - Comprehensive pattern matching
• Jailbreak Detection - Prompt injection protection
• Native Integration - Extends Agno's BaseGuardrail
• Async Support - Works with both run() and arun()

Quick Start

from agno.agent import Agent
from agno.models.openai import OpenAIChat
from sentinelseed.integrations.agno import SentinelGuardrail

agent = Agent(
    name="Safe Assistant",
    model=OpenAIChat(id="gpt-4o-mini"),
    pre_hooks=[SentinelGuardrail()],
)

response = agent.run("Hello! How can you help me?")

Configuration

SentinelGuardrail (Input)

from sentinelseed.integrations.agno import SentinelGuardrail

guardrail = SentinelGuardrail(
    seed_level="standard",       # minimal, standard, full
    block_on_failure=True,
    max_text_size=100000,
    validation_timeout=5.0,
    fail_closed=False,
    log_violations=True,
)

SentinelOutputGuardrail (Output)

from sentinelseed.integrations.agno import SentinelOutputGuardrail

guardrail = SentinelOutputGuardrail(
    seed_level="standard",
    max_text_size=100000,
    validation_timeout=5.0,
)

result = guardrail.validate_output(response.content)
if not result["safe"]:
    print(f"Output flagged: {result['concerns']}")

Safety Levels

Monitoring

Tracking Violations

guardrail = SentinelGuardrail(log_violations=True)

violations = guardrail.get_violations()
for v in violations:
    print(f"Risk: {v['risk_level']}, Concerns: {v['concerns']}")

Statistics

stats = guardrail.get_stats()
print(f"Total: {stats['total_validations']}")
print(f"Blocked: {stats['blocked_count']}")
print(f"Avg time: {stats['avg_validation_time_ms']:.2f}ms")

Combining Guardrails

from agno.guardrails import PIIDetectionGuardrail
from sentinelseed.integrations.agno import SentinelGuardrail

agent = Agent(
    name="Secure Agent",
    model=model,
    pre_hooks=[
        PIIDetectionGuardrail(),
        SentinelGuardrail(),
    ],
)

Validation Result

{
    "safe": bool,
    "should_proceed": bool,
    "concerns": list[str],
    "risk_level": str,  # low, medium, high, critical
    "gates": dict,
    "validation_time_ms": float,
}

Links

• Agno Documentation