The Pull-Based Agent Model

Vecta uses a short-lived, pull-based agent invocation: your server checks the control plane for work instead of accepting inbound connections.

What happens on a check-in

The agent runs from the machine's cron schedule and asks vectaapp.com what should happen. The control plane decides which jobs are due. The agent then processes those jobs locally, one at a time, and reports status, progress, counters, snapshot IDs, and a short error-message tail when a run fails.

The agent checks in every couple of minutes. A scheduled job therefore starts within about two minutes of its due time, rather than at an exact second. A first run for an enabled job that has never run also waits for the next check-in.

Note

Run now flags a job for the next check-in. It does not start a backup instantly. If the agent is offline, the flag waits until the agent reconnects.

Network consequences

Your server only needs outbound HTTPS access to vectaapp.com. You do not need to open an inbound port or expose an agent service to the internet.

What the agent receives

The agent receives non-secret job configuration: source path, destination URL, SSH port, and schedule. Destination credentials and the repository password are resolved locally on the agent.

What the agent reports

The dashboard can show job state and progress because the agent reports status, processed and transferred counters, duration information, snapshot IDs, exit codes, and short failure-message tails. It does not stream backup contents through the control plane.

What's next

See Job lifecycle & statuses for the states that these check-ins produce, or Agents for installation and cron files.