An Approval Inbox for AI Agents

Description and use

Implementing a unified approval inbox for AI actions eliminates scattered notifications, enables bulk processing, and provides auditability.

Problem

Managing approvals from multiple AI agents across channels leads to scattered notifications and lost tasks.

Outcome

A unified approval inbox reduces interruptions, improves tracking, and enables bulk decisions, while maintaining auditability and expiry controls.

Workflow

• 1. Agent calls POST /v1/actions to create an action card for each content item.
• 2. Cards appear in the shared inbox, tagged by kind and target URL.
• 3. Reviewer filters cards by kind or other criteria.
• 4. Reviewer uses POST /v1/actions/bulk-decision to approve or reject multiple actions at once.
• 5. Agent polls for status; upon approval, it proceeds to publish or next step.
• 6. Expired actions automatically move to expired state after expires_in.
• 7. All decisions are logged in the audit log for traceability.

Evidence

A single approval inbox for AI agents means every pending action — from every agent, on every channel — lands in one triage queue instead of scattered across chat pings you can lose track of.

Source

POST /v1/actions/bulk-decision lets a reviewer approve or reject a set of action IDs in one call, which is what the inbox UI's "select all matching" does under the hood.

Source

Every decision, whoever made it and whenever it happened, is recorded in the audit log — so "who approved the post that went out with the wrong CTA" is always answerable, individually decided or bulk.

Source