Getting Started with ArmorHealth

Adopting ArmorHealth has the same two independent pieces as any ArmorIQ vertical: the policy side (a bundle template applied to your agent) and the enforcement side (the agent's code actually calling through ArmorIQ before it touches claims, provider, or member data).

Prerequisites

• A verified ArmorIQ account and organization membership.
• An ArmorIQ API key - see API Keys.
• ArmorHealth enabled for your org (see below).
• Your health-domain agent registered as a target in ArmorIQ (targetType: agent).

Step 1: Enable ArmorHealth

ArmorHealth is an optional product toggle:

1. Go to Settings → Product Preferences.
2. Toggle ArmorHealth on for your org.
3. The ArmorHealth section (Dashboard and Policy Studio) appears in your sidebar.

Step 2: Apply a bundle template

Pick your agent - Member Care Navigator, Claims Accuracy Copilot, Benefit Suggestion Agent, or Risk Stratification Agent - and one of its bundle templates from the ArmorHealth Dashboard. Click Apply for an instant, ready-made policy, or Customize to open it in the Policy Studio first. See Bundle Templates for the full catalog and Concepts for what's inside the YAML.

Step 3: Wire the agent through Google ADK

ArmorHealth's health-domain agents are built on Google ADK (Agent Development Kit) and run in SDK mode - each tool call goes through ArmorIQ's plan → token → invoke flow via the ADK integration, not the MCP proxy. Instead of duplicating that integration here, see the dedicated Google ADK guide, which covers the full setup:

from google.adk.agents import Agent
from armoriq_sdk.integrations.google_adk import ArmorIQADK

armoriq = ArmorIQADK(api_key=os.environ["ARMORIQ_API_KEY"])

root_agent = Agent(name="member_care_navigator", model="gemini-2.0-flash", tools=[...])

async def handle_request(member_email: str, message: str):
    scope = armoriq.for_user(member_email, goal=message)
    scope.install(root_agent)
    try:
        async for event in runner.run_async(user_id=member_email, new_message=message):
            yield event
    finally:
        scope.uninstall(root_agent)

ArmorIQ hooks into ADK's after_model_callback (capture the plan and mint a token), before_tool_callback (enforce the policy - allow, hold, or block), and after_tool_callback (write the audit log). The health-specific pieces (PHI access control, provider network rules, claim processing thresholds) live entirely in the policy the bundle template deploys - nothing about the ADK wiring changes per agent type.

Step 4: Handle holds and delegated approvals

A held tool call (for example, propose_oon_referral or submit_clean_claim in a restricted bundle) opens a delegation request. The ADK bundle can auto-poll for a decision when the policy allows it - see Policy holds and delegation. ADK-based integrations get a longer client-side approval window (around 30 minutes) than direct-client integrations like ArmorPay's (around 5 minutes), which matters for workflows like site-of-care scheduling that may wait on a case manager rather than an instant approver.

Next steps