glossary · Running it

Unattended execution

definition

Unattended execution is the mode where a scheduled or triggered run executes with nobody at the keyboard: a dispatcher hands the work to an execution backend, and humans participate only at gates.

in one line: Nobody at the keyboard. People join only at gates.

in context

Where does unattended execution fit?

Same workflow. Nobody driving. In ConvOps, a schedule fires, the work goes to a runner, and people step in only at the checkpoints you placed.

schedule fires
dispatchto a runner
the steps runno one watching
sign-offapproval
receiptsession, cost
two ways to say it

What is unattended execution, in plain words?

Same idea at two depths: the plain version, then what the engine actually does.

It is not a second process. It is the same workflow with the person out of the driving seat.

Each run leaves a receipt: what ran, how many retries, what it cost.

for engineers

tasks_dispatch hands a task to an executor: an isolated run on Kubernetes, your own OpenCode server, or your own runner, which claims the work over HTTPS and reports back.

Dispatch is written in the same transaction as the task, and a reaper cleans up runs that die. Each run records session, retries and cost in USD.

the longer answer

Why does unattended execution matter?

Unattended is not a different kind of process; it is the same workflow with the human removed from the driving seat. That framing matters because it determines what you must build: not a second system, but a dispatch path (who runs the agent), a selection mechanism (what it works on), and reliability plumbing (what happens when the backend is down).

The reliability part is where naive implementations fail quietly. Dispatch must survive crashes between systems (written transactionally, redelivered idempotently), and a run that dies must be reaped rather than leak.

Judged well, unattended execution is measured in receipts: each run should record its session, its retries, and its cost, so autonomy shows up as a line item you can read rather than a bill you discover.

There is also the question of what a run does when it needs a person. A run with nobody watching cannot ask in the chat. The robust answer is that the process itself contains the stopping points: the run reaches a gate, that piece of work waits there, and a person answers when they can, while other work keeps moving. Without that, the run either guesses or stops for good.

in convops

How does unattended execution work in ConvOps?

In ConvOps, unattended runs start from a schedule or from dispatching a task to an executor: an isolated run with one Kubernetes pod per run, your own OpenCode server, or your own runner, which claims work over HTTPS. A schedule can spawn a new task on each fire or resume the same task at its current step. A run that fails on a model or infrastructure error resumes on its own; after three failures in a row the task stalls until a person dispatches it again. Each run records its session, retries, tokens and cost. A run that hits its time limit is retried by hand, with Resume or Start fresh. For a hands-on setup, see how to run Claude Code unattended on Kubernetes.

questions

What do people ask about unattended execution?

What is unattended execution in ConvOps?

Unattended execution is a run with nobody at the keyboard: a schedule or a dispatch hands the work to an execution backend, and people take part only at gates. In ConvOps that backend is an executor, such as an isolated run with one Kubernetes pod per run, your own OpenCode server or your own runner. The workflow, gates and record are the same as in attended work.

How do AI agents run unattended?

A trigger, such as a schedule or a dispatch, starts the agent in non-interactive mode on a runner. ConvOps supplies what the run needs around that: a workflow to follow step by step, gates where the work waits for a person, and a record kept outside the run. A run that fails on a model or infrastructure error resumes on its own, and after three failures in a row the task stalls until a person dispatches it again.

What happens when an unattended run needs a human?

That task waits and other work keeps moving. In ConvOps the run reaches a gate or marks its task blocked, and a person answers when they can, from any client. The run then continues from the same step, so the agent does not have to guess an answer nobody gave.

How do you keep the cost of unattended runs visible?

Record it per run and cap it per run. ConvOps records each run's session, retries, tokens and cost when the run ends, beside the metrics you define and bind for the work itself. In Claude Code print mode, the --max-budget-usd and --max-turns flags stop a single run at a ceiling you set.

Which teams move their work to unattended runs?

Teams whose process already runs well attended and whose volume or timing no longer suits a person at the keyboard, such as nightly checks, queued fixes or scheduled reports. ConvOps lets them move the same workflow to an executor, so people take part only at gates and every run leaves the same record as attended work.