# Channel A **channel** is a persistent communication pathway between an AI [agent](agent) and a human. In the context of AgentRQ, a channel is the real-time connection through which agents send [notifications](notification), receive replies, and maintain ongoing conversations throughout [task](task) execution. ## How Channels Work in AgentRQ When a [Claude Code](claude-code) agent connects to AgentRQ via [MCP](mcp), it gains access to a workspace-scoped communication channel. All messages, task updates, and human replies flow through this channel. The channel is identified by a unique workspace ID and is used in every MCP tool call — ensuring that messages are routed to the correct [workspace](workspace) and [task board](task-board). ## Channel Properties - **Persistent** — The channel remains open across multiple agent sessions - **Real-time** — Messages appear instantly on the task board - **Scoped** — Each workspace has its own isolated channel - **Threaded** — Each task has its own conversation thread within the channel ## Channels vs. Tasks | Concept | Description | |---------|-------------| | Channel | The connection between agent and workspace | | Task | A specific work item with its own conversation thread | | Message | A single exchange within a task's thread | A channel can carry many tasks. Each task is an independent conversation thread. Multiple agents can share a single channel if they're part of the same [workspace](workspace). ## SSE and Channel Transport AgentRQ uses [Server-Sent Events (SSE)](sse) as the transport layer for real-time channel delivery. This means agents can push messages to humans without polling — updates arrive immediately as the agent sends them. ## Related Terms - [Workspace](workspace) - [Task](task) - [Notification](notification) - [SSE](sse) - [Bidirectional Messaging](bidirectional-messaging)