Skip to main content
A channel lets people send requests to a Computer Agent from a service they already use. Channel connections belong to one selected Agent, even when several Agents share the same Computer.

​
Plan access before connecting

Decide:
  • which Agent should receive the messages
  • which workspace, account, or bot identity it should use
  • whether direct messages are allowed
  • which groups or channels are allowed
  • whether the Agent must be mentioned
  • who may request actions with external or destructive effects
Prefer a narrow allowlist. A shared workspace connection does not mean every channel should be able to invoke the Agent.

​
Connect and verify

Open the selected Agent’s Computer settings, choose a supported channel, and complete the displayed connection or sign-in flow. Some channels use credentials; others open a web or device-login flow. After connecting:
  1. Confirm the channel reports ready or connected.
  2. Send a message from an allowed location.
  3. Confirm the intended Agent receives it.
  4. Test mention behavior and a denied location.
  5. Confirm the reply appears only where expected.
Starting a login flow is not proof that the channel works. Finish the flow and test the actual inbound and outbound path.

​
Keep channel behavior useful

Channel instructions should define what the Agent owns, when it should ask for confirmation, and how it should report completed or blocked work. Proactive checks should remain quiet when nothing material changed. If several Agents share a Computer, configure and test each channel against its intended Agent account. Do not reuse a broad connection that routes unrelated customer or operational traffic to the wrong Agent. Disconnect unused channels and review allowlists when teams or responsibilities change.
Configuring channels through the API? See Computer Agent skills, credentials, and channels.