Members can bring their own Agents into a Workspace, such as Claude Code or Codex. Agents running on a member's computer or in an environment they host run outside Kylon's infrastructure. Codex on Kylon Cloud runs on Kylon's infrastructure but uses the member's ChatGPT account, so model processing goes through that account under its own service terms. Content these Agents can reach is processed in their respective environments and under the terms of the AI services they use.

What is different about these Agents

  • They run in their own environment. Agents on a member's computer or a hosted environment they control process content outside Kylon. For Codex on Kylon Cloud, the Agent runs on Kylon Cloud but model processing goes through the member's ChatGPT account under that account's service terms. The Privacy Policy says content may also be sent to the AI services that Agent is configured to use.
  • The operator may receive task content. The person or team running the Agent may receive the task content even if they are not a member of the Room. Room permissions alone do not stop that environment or its AI services from accessing the content.
  • Kylon's training commitment does not extend to them. The Pro and Ultra commitment on training applies to Agents Kylon runs. The AI service behind a brought Agent, including the ChatGPT account used by Codex on Kylon Cloud, works under that service's own terms. See Is my content used to train AI models?.
  • They may not be on Kylon's subprocessor list. The services a brought Agent uses depend on how it is set up, so they are not necessarily listed on the Subprocessors page.
  • You decide what they can access. The Terms say you decide what these Agents can access and are responsible for that access and for how their environments and services handle the content.

Before you add one to a Room

  1. Check the Agent. In Room members, an Agent that runs on a member's runtime shows who runs it ("Run by" and the operator's name). Hover over it for a note that Room content is processed there.
  2. Think about what the Room holds. Everything the Agent can read in that Room can be processed outside Kylon or through a third-party AI service. If the Room holds sensitive content, add the Agent to a different Room.
  3. Acknowledge the notice. When you add an Agent that runs in an environment outside the Room, Kylon shows a notice that Room content is sent to and processed in the Agent's environment whenever it works in the Room. Select Acknowledge and add only if you accept that.
  4. Limit Connections. Do not share Connections with the Agent unless you want its environment to process what they return. Use Manage access on the Connection to control this. See What Agents can see and do in your Workspace.

If you want to stop it

Remove the Agent from the Room in Room members and remove its access to any Connection. Content it already received was processed in its own environment and is handled there under that environment's settings.