# Notification A **notification** is a real-time message sent by an AI [agent](agent) to a human, informing them of progress, blockers, or completed work. Unlike an [approval](approval) request that blocks the agent until the human responds, a notification is typically informational — the agent may continue working while the human reviews it. ## Purpose of Notifications Notifications serve several roles in agent workflows: - **Progress updates** — "I've finished the refactor, running tests now" - **Completion signals** — "All 47 tests are passing, PR is ready for review" - **Blocking alerts** — "I encountered a permission error and need your help" - **Decision requests** — "I found two approaches — which do you prefer?" They keep humans informed without requiring constant manual check-ins. ## Notifications vs. Approvals | | Notification | Approval | |--|-------------|---------| | Agent behavior | Continues working | Waits for response | | Human action required | Optional | Required | | Use case | FYI updates | Permission gates | ## Notifications in AgentRQ AgentRQ delivers agent notifications in real time to your [task board](task-board) via [SSE](sse). When a [Claude Code](claude-code) agent calls the `reply` MCP tool, the message appears instantly — no polling, no delays. You can configure notifications per [workspace](workspace) and receive them on any device that's connected to your AgentRQ session. ## Designing Good Notifications - **Be specific** — "Tests failed: 3 failures in auth module" beats "something went wrong" - **Include context** — Tell the human what the agent was doing when the notification fired - **Signal what's needed** — Make it clear whether a response is required or it's just informational - **Don't over-notify** — Too many notifications train humans to ignore them ## Related Terms - [Approval](approval) - [Bidirectional Messaging](bidirectional-messaging) - [Task](task) - [SSE](sse) - [Human-in-the-Loop](human-in-the-loop)