# Bidirectional Messaging **Bidirectional messaging** is the ability for both an AI [agent](agent) and a human to exchange messages in a shared conversation channel. Unlike a one-way notification system where the agent simply alerts the human, bidirectional messaging allows the human to reply, ask questions, provide corrections, and guide the agent — and for the agent to incorporate that feedback in real time. ## Why Bidirectionality Matters Pure autonomous execution has a fundamental problem: the agent can't ask for help. When it hits an ambiguous requirement, encounters an unexpected error, or needs a decision it can't make alone, it either guesses or stops entirely. Bidirectional messaging solves this by creating a real-time feedback loop: 1. Agent encounters ambiguity or needs approval 2. Agent sends a message to the human via [AgentRQ](/) 3. Human responds with guidance, corrections, or approval 4. Agent incorporates the response and continues working ## Bidirectional Messaging in AgentRQ AgentRQ implements bidirectional messaging over [MCP](mcp) using two complementary tools: - **`reply`** — The agent sends a message to the human on a specific task's chat - **`getTaskMessages`** — The agent retrieves the conversation history, including human replies This creates a persistent, thread-like conversation attached to each [task](task), visible on the [task board](task-board). ## Comparison with One-Way Notifications | Feature | One-Way Notification | Bidirectional Messaging | |---------|---------------------|------------------------| | Agent → Human | Yes | Yes | | Human → Agent | No | Yes | | Real-time feedback | No | Yes | | Agent can ask questions | No | Yes | | Supports approvals | No | Yes | ## Related Terms - [Task](task) - [Notification](notification) - [Approval](approval) - [Channel](channel) - [MCP Tool](mcp-tool)