Audit your AI agents before they reach production
The Tvijo AI Ops Starter Kit checks whether every agent declares the controls an operator needs: ownership, data boundaries, approval rules, budget limits, a kill switch, and an audit trail.
$ ai-ops-kit audit --registry agents.yaml
CRITICAL support-agent: kill switch is missing
HIGH support-agent: monthly budget is not declared
HIGH support-agent: approval policy is missing
Report written to audit-report.json
AI agent readiness audit
Seven declarations that expose expensive gaps early
A small registry gives engineering, security, and operations teams one reviewable record before an agent receives production access.
Named owner
Someone is accountable for changes, incidents, and retirement.
Purpose and status
The intended job and lifecycle state are explicit.
Data classification
The registry declares what information the agent may handle.
Human approval
High-impact actions have a documented approval policy.
Budget limit
Spend is bounded before loops or retries become costly.
Kill switch
Operators know how to stop the agent quickly.
Audit log
The declaration identifies where evidence should be recorded.
From inventory to review
A three-step, local workflow
Register
Describe agents in a version-controlled YAML registry.
Audit
Run one local command in a laptop workflow or CI pipeline.
Review
Share the text or JSON report and assign remediation owners.
What this audit does — and what it does not
The starter kit validates declared governance controls. It does not observe runtime behavior, verify that a kill switch works, certify security or legal compliance, or prove production readiness. Treat the report as a structured starting point for evidence collection and human review.
Start with the open source baseline
Clone the repository, adapt the sample registry, and keep the resulting evidence in your own environment. The CLI sends no telemetry and makes no network calls.
Need runtime controls after the declaration audit?
Tvijo AI OS adds the operating layer around model routing, budgets, approvals, policy, and audit evidence. Bring the starter-kit findings to a focused readiness review.
Request an AI readiness review