Get started

Schedules · act, escalate, defer

Autonomy with a spine.

Letting agents run on schedules is easy. Deciding when they should not is the actual product. Before each scheduled run, Quox checks budget, agent standing, tool scope and risk, retry and failure history, and DEFCON, in order, and lands on one of three decisions: act, escalate to a human, or defer. No vibes, no risk scores: gates, in order, on the record.

Included in QuoxCOREFree, self-hosted, nothing to buy.
Decisions3Gate orderfixedEscalation targetyour inbox

In plain words

What it is, where it lives, when to reach for it

What is it
A decision layer in QuoxCORE that gates every scheduled agent run: act, escalate to a human, or defer.
Where do I use it
In the dashboard: schedules live on the Triggers page, and escalations arrive in your Inbox.
When would I use it
When agents run on schedules without you watching and some runs should not go ahead.
How do I use it
You do not run the gates; schedule an agent on the Triggers page and every run is checked automatically.

QuoxCORE is the free, self-hosted platform underneath this. What is QuoxCORE

The vocabulary

Act. Escalate. Defer.

actAct

The gates are clear: the run dispatches with its own token, tool-call and runs-per-day budget.

▸ dispatch · budgeted

escalateEscalate

Something needs judgement: by default a budget warning or a high-risk tool in the plan. The run parks as pending approval and a human gets it in the inbox.

▸ pending_approval · inbox

deferDefer

The retry limit is reached. The run is recorded as deferred with its reason, rather than burning money to look busy. A spent daily budget skips the run before it starts.

▸ deferred · reason recorded

The order matters

The gates, in order.

01

Budget exhausted

The run is skipped before it starts and logged. It is not recorded as a decision.

▸ skip
02

Budget warning

Escalates by default. The escalation policy on a schedule can change that.

▸ escalate
03

High-risk tool in the plan

SSH execution, system administration, extended docker control, VM actions and security audits are on the high-risk list by default, and the escalation policy decides what happens when one appears in a plan.

▸ policy decides
04

Retry limit reached

▸ defer
05

Schedule close to its failure cap

▸ escalate

First gate that fires wins. Agent standing, tool scope, loop-mode and DEFCON checks also sit in this sequence.

The routine fleet

Agents on a rota, on a leash.

Cron schedules with timezones, a task template, an escalation policy and a budget per run. Three consecutive failures stands a schedule down by default. Agents ship with suggested cadences and budgets, for example six-hourly on fifty thousand tokens for the security agent and every fifteen minutes on thirty thousand for the network metrics agent. These are declared defaults: you create the schedule that uses them. The orchestrator's manifest marks it as not autonomous, though schedules are not yet blocked on that flag.

cron + timezone · budget per run · failure caps

Triggers

Plain words in, cron out.

Describe when in plain language and Quox writes the schedule: triggers carry a calendar view as well as a list, so the fleet's week reads like a diary. Pause, resume, search and attribute every trigger to the agent that owns it.

nl to cron · calendar and list views

A dedicated CalendarQ surface for the whole fleet's diary is on the roadmap. ▸ coming · calendarq

Trigger Management on the Schedules tab: fifteen active schedules with their workflow, the cron expression in plain words and in cron form, next run countdown, timezone, and pause and open actions

The handoff

Escalations land on a person.

An escalated run files an approval into the inbox, assigned to the schedule's owner, with approve and deny actions. The autonomy framework does not have a second inbox: it uses the one you already work.

kind approval · assigned to owner

Every decision

Deciding is an event.

Each decision made for a run writes a decision row and an evidence envelope with the reason. Runs skipped before the decision (daily budget spent, kill switch) are logged but not recorded as decisions. Forty-one tests across four suites cover the decision engine.

autonomy_decisions · quox.autonomy.decision.*

Licence-tiered authority ceilings are built and tested but not yet wired into scheduled runs; they arrive with licence enforcement. ▸ coming · authority levels

Ready when you are

Let them run. On your terms.

The autonomy framework ships with QuoxCORE: schedules, gates and escalation included.