SDK Quick Start — Insurance: P&C, Life & Annuity

Getting started: Insurance: P&C, Life & Annuity

Control what your AI can do. The core SDK checks policy before your workflow runs an action and again before its result is released. Denied actions stop. Actions that need human review stay paused until a signed outcome arrives.

Catalog context. For a claims workflow, capture the claim stage and the reason for routing it to a reviewer. Keep the review outcome separate from the action the application later takes. Determine applicable requirements by jurisdiction, insurance line, and intended use. A recorded review or bias-test result does not establish that the process meets those requirements.

Use industry fields with the core SDK

Industry vocabularies supply typed values that the SDK sends as policy fields and evidence attributes with one governed call. Choose values that describe the actual workflow. Adding a vocabulary does not activate rules or assign regulatory requirements.

If you need the cross-industry baseline first, open the core SDK reference, Python core, or .NET core before continuing here.

Step 1 — Install the approved packages

In the customer portal, open Tools and SDKs. Select your language and adapters, then follow the installation instructions shown there.

VeriProof SDKs are distributed through customer-gated GitLab packages, not public registries. See SDK access for workstation and CI guidance.

VeriProof packages used here: @veriproof/sdk-core (TypeScript), veriproof-sdk (Python), Veriproof.Sdk.Core (.NET).

Step 2 — Configure

Store the application SDK key as VERIPROOF_API_KEY and create one client when the application starts. Publish the policy you want to enforce, then confirm its field contract against the facts and industry values your application supplies.

Copy the application ID and key from Set up SDK. The application ID is not its display name. Try a live demo  to see a governed workflow, or request an evaluation  if you need a workspace.

Step 3 — Govern the action

Put the protected business action inside govern. The SDK checks policy before it calls the action and before it releases the result. Deny and human-review decisions stop execution here.

import { Veriproof } from '@veriproof/sdk-core';
import type { GovernedResult, IndustryFact, JsonValue } from '@veriproof/sdk-core';

const veriproof = Veriproof.fromEnv();

export async function runProtectedAction<TInput extends JsonValue, TValue>(
  request: TInput,
  action: (input: TInput, signal: AbortSignal) => Promise<TValue>,
): Promise<GovernedResult<TValue, TInput>> {
  // Catalog values for this vertical. Replace them with facts from the current workflow.
  const industry: IndustryFact[] = [
    { attribute: "insurance.policy.lifecycle_status", policyInput: "insurance_policy_lifecycle_status", value: "application_received" },
    { attribute: "insurance.policy.transaction_type", policyInput: "insurance_policy_transaction_type", value: "new_business" },
    { attribute: "insurance.policy.line_of_business", policyInput: "insurance_policy_line_of_business", value: "personal_auto" },
  ];

  // Policy runs before action() and again before the result is released.
  // A deny or review decision means action() never runs.
  return veriproof.govern(
    "insurance.pc.life.annuity.evaluate",
    request,
    ({ input, signal }) => action(input, signal),
    { industry },
  );
}

Only completed, completed_with_warning, and transformed results carry a released value. Handle every other status with its safe message, reason code, and evidence reference. For review decisions, save the checkpoint and continue through the human review workflow.

Add typed industry values

Each core SDK ships this vertical's vocabulary as generated types. A typed value becomes one policy field and one evidence attribute when you pass it to the governed call. Use application facts for values that are specific to your workflow.

Find this workflow's vocabulary in the Insurance: P&C, Life & Annuity reference. Use the attribute names below and verify their policy-field mapping before using them in a rule. Sample values show the format; they are not recommended outcomes for your application.

TypeScript

import { Industry } from '@veriproof/sdk-core';
import { PolicyLifecycleStatus, PolicyLifecycleStatusMeta, PolicyTransactionType, PolicyTransactionTypeMeta, InsuranceLineOfBusiness, InsuranceLineOfBusinessMeta } from '@veriproof/sdk-core/industries/insurance';

// Replace sample values with facts from the current workflow.
// Each value becomes one policy field and one evidence attribute.
const industry = [
  Industry.fact(PolicyLifecycleStatusMeta, PolicyLifecycleStatus.ApplicationReceived),
  Industry.fact(PolicyTransactionTypeMeta, PolicyTransactionType.NewBusiness),
  Industry.fact(InsuranceLineOfBusinessMeta, InsuranceLineOfBusiness.PersonalAuto),
];

const result = await veriproof.govern(actionName, request, runWorkflow, { industry });

Python

from veriproof_sdk.verticals.insurance import PolicyLifecycleStatus, PolicyTransactionType, InsuranceLineOfBusiness

# Replace sample values with facts from the current workflow.
# Each value becomes one policy field and one evidence attribute.
industry = (
    PolicyLifecycleStatus.application_received.fact(),
    PolicyTransactionType.new_business.fact(),
    InsuranceLineOfBusiness.personal_auto.fact(),
)

result = await vp.govern(action_name, request, run_workflow, industry=industry)

.NET

using Veriproof.Sdk.Core.Industries.Insurance;

// Replace sample values with facts from the current workflow.
// Each value becomes one policy field and one evidence attribute.
var options = new GovernOptions
{
    Industry =
    [
        PolicyLifecycleStatus.ApplicationReceived.Fact(),
        PolicyTransactionType.NewBusiness.Fact(),
        InsuranceLineOfBusiness.PersonalAuto.Fact(),
    ],
};

var result = await veriproof.GovernAsync(actionName, request, RunWorkflowAsync, options);

Workflow examples from the catalog

• Underwriting, Rating & Pricing: Rating Factor Category. OPA rule: deny any rating action where factor_category is prohibited_characteristic_proxy or social_media_derived until Colorado Reg 10-1-1 bias test has been completed and approved. (Colorado Sb21 169: ECDIS factors require bias testing for proxy discrimination)
• Underwriting, Rating & Pricing: ECDISBias Test Outcome. OPA rule: block production deployment of any underwriting model where ecdis_bias_test_outcome is testing_not_yet_performed, material_disparity_corrective_action_pending, or model_suspended_pending_remediation. (Colorado Reg 10 1 1: Annual compliance report due Dec 1 (life); July 1, 2026 (auto and health); bias test outcomes must be documented)
• Underwriting, Rating & Pricing: Insurance AIModel Inventory Status. OPA rule: deny any agentic underwriting or claims decision where the dependent model status is not_in_inventory_shadow_model or third_party_vendor_model_pending_assessment. (Naic Model Bulletin: AIS Program must maintain inventory of all AI systems used in regulated decisions)
• Claims Intake, Triage & FNOL: Claim Severity Tier. OPA rule: route all tier_7 and above claims to human adjuster, block STP. Require HITL with senior_adjuster role for tier_8 and above. Require claims_compliance_officer for tier_10.

Next steps

• Open the full Insurance: P&C, Life & Annuity governance reference for all 31 enum categories, policy patterns, and source standards.
• Review TypeScript SDK core for stop handling, human review continuation, and the OpenAI Agents adapter.
• Review Python SDK core for stop handling, human review continuation, and the Microsoft Agent Framework adapter.
• Review .NET SDK core for stop handling, human review continuation, and the Microsoft Agent Framework adapter.
• Browse all vertical references to understand how the core governance vocabulary layers underneath vertical-specific attribute sets.
• Compare this guide with the public industry pages  for a business overview of the same workflows.