Integrate Sentinel in 5 Minutes: A Quick Start Guide

You have an AI agent. It calls tools, queries databases, maybe even moves money. You want a safety layer that validates decisions before they become actions.

Here's how to add Sentinel to your stack — four ways, all under 5 minutes.

Option 1: Python SDK (pip install)

The fastest path. Works with any Python agent.

pip install sentinelseed

from sentinelseed import Sentinel

sentinel = Sentinel(seed_level="standard")

Validate user input before sending to your LLM

result = sentinel.validate_input(user_message)

if not result.is_safe:

print(f"Blocked: {result.attack_types}")

Don't send to LLM

Validate LLM output before executing actions

result = sentinel.validate_output(llm_response, user_message)

if not result.is_safe:

print(f"Seed failed: {result.gates_failed}")

Don't execute

That's the core loop: validate input, run your agent, validate output.

The heuristic layer (700+ patterns) runs locally with zero API calls. For deeper analysis, enable the semantic layer with any LLM provider:

sentinel = Sentinel(

seed_level="standard",

semantic_provider="openai", # or "anthropic", "openai_compatible"

semantic_api_key="sk-...",

)

Option 2: TypeScript SDK (npm install)

Same concept, TypeScript-native.

npm install @sentinelseed/core

import { validateTHSP } from '@sentinelseed/core';

const result = validateTHSP(userMessage);

if (!result.is_safe) {

console.log('Blocked:', result.violations);

}

The TypeScript SDK includes the full THSP heuristic engine, HARM_PATTERNS for SQL injection detection, and refusal detection utilities.

Option 3: Framework Integration (LangChain, CrewAI, etc.)

If you're using a popular agent framework, Sentinel has native integrations:

LangChain

from sentinelseed.integrations.langchain import SentinelGuard

guard = SentinelGuard()

safe_chain = guard.wrap(your_chain)

Every chain invocation now runs through THSP validation

result = safe_chain.invoke({"input": user_message})

CrewAI

from sentinelseed.integrations.crewai import SentinelCrew

crew = SentinelCrew(agents=[...], tasks=[...])

All agent actions validated automatically

LangGraph

from sentinelseed.integrations.langgraph import add_sentinel_node

graph = add_sentinel_node(graph, before="agent")

Safety checkpoint added to your graph

We support 30+ frameworks including DSPy, LlamaIndex, AutoGPT, Letta, Agno, OpenAI Agents SDK, Anthropic SDK, Solana Agent Kit, and more. Full list at [sentinelseed.dev/integrations](/integrations).

Option 4: Visual Builder (No Code)

Don't want to write code? Use the platform:

5. Add your LLM API key (BYOK — keys are encrypted client-side)

6. Test in sandbox mode

7. Deploy with one click

Your agent gets a live endpoint with Sentinel protection built in.

What Gets Validated?

Every validation checks four gates (the THSP Protocol):

All four must pass. The absence of harm is not enough — there must be genuine purpose.

Validation Layers

The SDK runs up to four layers of protection:

You can enable or disable each layer based on your security needs and cost budget.

Cost

The heuristic layer (L1 + L3) is completely free and runs offline. No API calls, no data leaving your system.

The semantic layer and L4 Observer use LLM calls — you choose the provider and model. With gpt-4o-mini, expect around $0.0005 per validation.

On the platform, execution costs $0.003 per run, paid with credits (SOL, USDC, or $SENTINEL with a 20% bonus).

Next Steps

Pick the integration that fits your stack and start validating.

The Sentinel Team