Cloud workers in Remote Control

Owners configure an organization compute cap, a cumulative standing allowance, designated API keys and approved encrypted vault pools. Machines must be enabled, and the deployment must supply a reviewed private worker image and provider token. The image and inference, Hawkeye enrollment and verifier adapters still need integration; this implementation has not been deployed.

In Remote Control, choose a size, region, lifetime and named pool. Review the live hourly/monthly price and maximum lifetime reservation before approving creation. An agent uses worker_quote then worker_spawn; it needs a standing allowance or an owner must approve that exact quote using its review link. worker_list shows state, estimated cost and recent commands. worker_command runs exec or PTY work through existing machine grants and signed results. Queued receipts are explicitly unverified; use the existing machine command status to read the agent outcome.

The terminal grid replays ordered output, labels panes with their agent, and starts read-only. Choose Take input to obtain the existing control lease. worker_action supports stop, start and destroy. Destroy requires the exact worker ID confirmation. Stop remains billable. Start rotates the machine identity and retains the original expiry. The minute sweep destroys idle or expired boxes and retries provider errors; the guest also stops command processes at expiry. Provider outages can delay deletion and compute billing. An open command or PTY counts as busy for idle detection.

REST uses GET /v1/orgs/{orgId}/workers, PUT /v1/orgs/{orgId}/workers/policy, POST /v1/orgs/{orgId}/workers/quotes, GET /v1/orgs/{orgId}/workers/quotes/{quoteId}, POST /v1/orgs/{orgId}/workers, POST /v1/orgs/{orgId}/workers/{workerId}/commands, and POST /v1/orgs/{orgId}/workers/{workerId}/actions.

We use essential cookies to keep you signed in. See our Privacy Policy for details.