Build an Agent with IRIS
Follow these steps to register an agent, set up governance files, and prepare for runtime enforcement.
1. Install IRIS
iris quickstart
Requires Python 3.10+. No cloud account. Runs fully local.
2. Discover existing agents (optional)
Before registering, scan your repo for ungoverned LLM calls and agent patterns.
iris scan --discover --govern # show one-line fixes
3. Register your agent
Registration creates an AgentPassport — the inventory record regulators and security teams expect. Primary command: iris declare (iris register is an alias).
--name loan-processor \
--owner you@company.com \
--team platform \
--compliance colorado-ai-act \
--high-risk
Creates governance/agents/loan-processor/passport.yaml.
4. Check compliance posture
iris compliance check --framework colorado-ai-act
IRIS shows which rules you violate and how to fix each one in plain English.
5. Write policy intent
Edit governance/agents/loan-processor/policy-intent.md:
It may call the credit bureau API in US regions only.
User consent must be logged before any loan decision.
It must never call any API not listed here.
6. Compile to Cedar
iris compile --agent loan-processor
iris policy diff --agent loan-processor # preview drift
iris preview --agent loan-processor # risk impact
Commit both policy-intent.md and policy.cedar to Git. Review in a PR.
7. Wire runtime enforcement
See Runtime Enforcement for drop-in LLM clients and @agent.guard() decorators.
Tip: Check governance/ into your repo. Validate on every PR with iris scan --fail-on critical and iris compliance check.