Your AI Transformation, Enterprise Ready
The agentic integration platform that enables zero-trust AI systems to do real work. Connect AI to critical systems with enterprise-grade safety and approval workflows.
Product Features
Enterprise-grade AI safety and control, built for scale
Agent Gateway
Unified, Enterprise-Grade Interface for LLMs. Connect your enterprise to any model, local or cloud through our platform.
Drop-in replacement for OpenAI's API. Connect to any LLM through a single, enterprise-grade endpoint.
from openai import OpenAI
client = OpenAI(
base_url="https://guardrails.<YOUR_COMPANY>.com/api/v1",
api_key="<GUARDRAILS_API_KEY>",
)
response = client.chat.completions.create(
model="gpt-4",
messages=[{"role": "user", "content": "Hello!"}]
)- OpenAI-compatible API
- Multi-provider routing
- Enterprise controls
- Usage monitoring
Agent Portal
Bring LLMs into your workflow, wherever you are. Available on Discord, Slack, Teams, and custom integrations.
Audit Logs
Enterprise-grade compliance with customizable, strongly-typed, exportable audit logging for all user and AI interactions.
Enterprise-grade compliance with comprehensive audit logging. Track every AI interaction, approval decision, and system event with strongly-typed, exportable logs.
- Immutable audit trail
- Customizable log schemas
- Real-time streaming
- Compliance reporting
Directory Sync
Automated user and group provisioning for approval workflows. SCIM, SSO, and enterprise directory integration.
Seamlessly integrate with your existing identity infrastructure. Automated user and group provisioning ensures your approval workflows stay in sync with your organization structure.
- SCIM 2.0 protocol support
- Active Directory integration
- Real-time user provisioning
- Group-based permissions
Flexible Policy Engine
Define organizational policies for high-risk actions in natural language. Automatically convert to deterministic rules that evaluate every AI interaction consistently.
Write policies like "Travel expenses over $1000 require manager approval", "Production database access needs security team sign-off", or "Alert a manager if an individual is consistently using much more budget than other employees" in plain English. The engine converts these into deterministic rules that evaluate every accounts payable, travel, data access, and other high-risk AI interactions consistently across your organization.
- Natural language policy authoring
- Deterministic rule conversion
- High-risk action coverage
- Consistent evaluation across interactions
What is GuardRails ?
An agentic integration platform that sits between AI and your high-risk workloads, providing just-in-time feedback and approvals for critical AI operations.
Zero-Trust Architecture
Every AI action requires explicit approval before execution
Real-Time Controls
Dynamic policy enforcement with instant feedback loops
Secure Integrations
Connect AI to critical systems safely with built-in safeguards
Model Context Protocol
Native MCP server integration for seamless AI tool connectivity
from guardrails_agent import GuardRailsAgent, NotarizedApproval
from langchain.agents import AgentExecutor
# NotarizedApproval data structure
@dataclass
class NotarizedApproval:
approval_id: str
action: str
parameters: Dict[str, Any]
approver: str
timestamp: datetime
signature: str # Cryptographic signature
granted: bool = True
# Initialize LangChain agent with approval tools
agent = GuardRailsAgent(openai_api_key="your-key")
# Agent automatically follows approval workflow:
# 1. get_approval_dryrun - analyze requirements
# 2. get_approval - request actual approval
# 3. execute - run with notarized_approval
user_request = "Transfer $5000 to vendor ABC Corp"
response = await agent.run(user_request)
# Agent ensures execute tool ALWAYS has notarized_approval:
# {
# "action": "payment",
# "parameters": {"amount": 5000, "recipient": "ABC Corp"},
# "notarized_approval": {
# "approval_id": "uuid-here",
# "approver": "jane.doe@company.com",
# "signature": "crypto-signature",
# "granted": true
# }
# }Add human-language safeguards into your AI applications
Define safety policies in plain English and automatically validate AI responses before they reach your users. Catch potential issues and escalate gracefully.
from guardrails import validate_response
async def handle_customer_query(query, ai_response):
# Validate AI response against safety policies
validation = await validate_response(
query=query,
response=ai_response,
policies=[
"Always use polite and empathetic language",
"Avoid technical jargon, explain in user-friendly terms",
"Never make promises about features or timelines",
"Show understanding of customer frustration",
"Frame solutions in the customer's language"
]
)
if validation.safe:
return ai_response
else:
# Escalate to human with graceful message
await escalate_to_human(query, validation.reason)
return "I'm sorry, this response would violate my safeguards. Escalating to a human representative now."Response Validation
Check AI responses against safety policies before delivery
Graceful Escalation
Automatically route to human agents when policies are violated
Natural Language Policies
Write safety rules in plain English, not complex code
Historical Evaluation
Test policy changes against historical data before deployment
Reduce Risk
Eliminate unauthorized AI actions with mandatory approval workflows for high-risk operations.
Maintain Compliance
Built-in audit trails and compliance reporting for SOC 2, GDPR, and industry standards.
Scale Safely
Deploy AI at enterprise scale with confidence, knowing every action is controlled and monitored.
Intent Execution with Approval Workflow
Build durable, long-lived, high-risk AI workflows that survive failures and maintain state across complex enterprise operations.
Dryrun Analysis
Agent analyzes intent and determines approval requirements
Approval Request
System routes to appropriate approvers based on risk level
Secure Execution
Action executes only with valid notarized approval
Ready to secure your AI?
Join leading enterprises using Guardrails to deploy AI safely at scale.