Quickstart

Prerequisites

• A SinzAI account (request early access on the homepage)
• At least one AI agent running in your infrastructure (OpenAI, Anthropic, LangChain, CrewAI, etc.)
• Admin access to your API gateway or agent runtime for integration
• Optional: SSO provider (Okta, Azure AD, Google Workspace) for agent identity discovery

Step 1: Connect Your Infrastructure

SinzAI integrates with your existing infrastructure through the SinzAI Gateway — a lightweight proxy that sits between your agents and their target systems. Deploy the gateway as a sidecar, a reverse proxy, or an SDK middleware:

# docker-compose.yml
services:
  sinzai-gateway:
    image: sinzai/gateway:latest
    environment:
      SINZAI_API_KEY: ${SINZAI_API_KEY}
      SINZAI_TEAM_ID: ${SINZAI_TEAM_ID}
    ports:
      - "8080:8080"
    volumes:
      - ./sinzai-policies:/etc/sinzai/policies

// TypeScript / Node.js
import { SinzaiGateway } from "@sinzai/gateway";

const gateway = new SinzaiGateway({
  apiKey: process.env.SINZAI_API_KEY,
  teamId: process.env.SINZAI_TEAM_ID,
});

// Wrap your agent's execution
const governedAgent = gateway.wrap(myAgent);

Step 2: Discover Your Agents

Once the gateway is deployed, SinzAI automatically begins detecting agent traffic. Navigate to the SinzAI console → Agents to see your agent inventory:

1. 1Gateway traffic automatically surfaces agents making API calls
2. 2Connect your SSO provider (Settings → Integrations → SSO) to discover agent service accounts
3. 3Connect your cloud provider (Settings → Integrations → Cloud) to detect AI-related spend patterns
4. 4Review the Shadow AI panel — any unregistered agents appear here with a warning badge

Step 3: Register Your First Agent

For each agent in your inventory, complete the registration form:

1. 1Name & Team — Give the agent a descriptive name and assign it to a team
2. 2Registered Purpose — Write a clear statement of what the agent is designed to do (e.g., 'Sales lead qualification — researching prospects and drafting personalized outreach sequences')
3. 3Data Scope — Select the data stores the agent needs access to. Start minimal — you can expand later.
4. 4HITL Gates — Configure which actions require human approval. We recommend starting with: email sends, API writes, and any action on PII data stores.
5. 5Credential TTL — Set the credential lifetime. Default is 24 hours. Shorter for high-risk agents.

Step 4: Configure Your First Policy

Navigate to Policies → New Policy. Create a policy that defines what happens when guardrails are tripped:

policy "pii-access-control" {
  description = "Require HITL approval for PII access"
  type        = "access"

  match {
    data_classification = ["PII", "PHI", "PCI"]
    action_type        = ["read", "write", "delete"]
  }

  enforce {
    hitl_gate    = true
    approval_ttl = "10m"
    on_timeout   = "deny"
  }

  on_violation {
    action   = "halt"
    notify   = ["#security-alerts"]
    severity = "high"
  }
}

Step 5: Switch to Enforcement Mode

After validating your configuration in monitor mode, switch to enforcement mode:

1. 1Review the monitor-mode ledger entries to confirm no false positives
2. 2Navigate to Settings → Enforcement → Enable for your team
3. 3SinzAI now blocks violations in real time according to your policies
4. 4Monitor the Dashboard KPI cards for containment events, gate activations, and trust score changes