Persistent execution jobs
Start builds, tests and deploys asynchronously. Stream stdout and stderr, cancel work, retry failed attempts and keep results after reconnects.
Captain Relay connects AI assistants to enrolled computers through a policy-controlled, auditable relay — with human approvals when an action needs them.
Give an AI the narrow capability it needs — files, commands, long-running jobs or native desktop input — instead of exposing an unrestricted machine.
Start builds, tests and deploys asynchronously. Stream stdout and stderr, cancel work, retry failed attempts and keep results after reconnects.
On supported desktops, Captain Relay can capture the screen, discover windows and send precise pointer or keyboard input through native desktop APIs.
Allow, deny or require approval per capability and device. Reading a file does not have to grant shell or desktop control.
Captain Relay sits between the assistant and the computer, turning broad intent into a specific, attributable action.
The user signs in and the client proves its identity before the relay accepts a request.
The server checks device ownership and the exact policy for the requested capability.
When policy requires it, execution pauses until a human explicitly approves the action.
The enrolled agent performs the narrow action and sends the result back through the relay.
Captain Relay is designed so remote capability is explicit, scoped and revocable. The assistant cannot turn one granted tool into every other tool.
Enroll a device, choose the policies, then let your assistant do useful work through Captain Relay.
Open Captain Relay