v2.5

A new approval system lets you require human sign-off before agents execute sensitive actions. Using the @approval decorator alongside a HITL primitive (requires_confirmation, requires_user_input, or external_execution), you can define blocking or audit-mode approval gates with persistent database records for review and governance.

This means high-stakes operations—such as financial transactions, data modifications, or external API calls—don't proceed without explicit admin authorization. Every approval decision is recorded, creating a clear audit trail for compliance and operational review.

Details

• Blocking mode (@approval or @approval(type="required")) pauses execution and writes a pending record to your database. The run only resumes once an admin resolves it via db.update_approval(...) and agent.continue_run() is called.
• Audit mode (@approval(type="audit")) is non-blocking — the run continues immediately after the HITL interaction is resolved, while an audit log is created for compliance and traceability.
• Persistent approval records are stored in your configured database for compliance, audit, and post-incident review.
• Programmatic resolution is handled through your DB provider, with expected_status checks to prevent race conditions, and can be integrated into existing review workflows and dashboards.

Learn more in the Agno approval docs