Automations
Work starts where your team already works.
A label, a transition, a mention, a schedule or a call — each bound to an automation with a budget and a level of oversight.
What starts work
Six ways in. Fifteen kinds of start.
Exactly the kinds the control plane accepts, scoped to the repositories, projects and teams you choose. Nothing starts that you did not bind to an automation.
A label
Label an issue and a change opens against it, in the repository you set.
GitHub · Jira · LinearA transition
Move a Jira issue into the column you chose. The project is the scope, and the webhook renews itself before it expires.
JiraAn assignment
Assign a Linear issue to the app, scoped by team and project.
LinearA mention
Mention @devbyte in a comment to start work, or to steer work already running. A pull request marked ready for review starts a review.
GitHub · Jira · Linear · SlackA schedule
Run an automation on a schedule — the dependency sweep, the nightly review.
cronAn API call
Start a change from your own systems, with the same scope and the same automation.
REST
An automation binds what starts work to a workflow, a risk level and a level of oversight — trivial, simple or full, chosen at triage from the change’s own signals; a ticket that needs answers first gets them asked on the ticket, and parked work carries a date to come back; and the risk is computed from the change’s own signals, never set by hand.
Autonomy
Attended first. Unattended, earned.
A new automation runs with a person watching. It runs unattended only after its own record, at that risk level, has passed the required number of times. There is no switch to flip.
- AttendedA person watches every stage and signs off each one.
- SupervisedAfter clean runs, sign-offs move to the points the risk level requires.
- UnattendedEarned per automation and per risk level. Lost again the moment the record says so.
The line improves itself. You approve each change to it.
Every run is scored daily and failures are clustered by cause. A proposed fix — to a check, a route, a skill, a knowledge entry — is replayed against the runs that failed and against a set of known-good runs before you see it, then rolled out to a share of new work and rolled back on its own if the failure returns.
Budgets
Dollars, not tokens. A ceiling above every run.
Budgets are set in dollars — a cap for the organisation, the same ceiling for every member so nobody’s share shrinks when someone joins, and a budget on every change an automation starts. Tokens are derived for display. Every stage reports what it spent, rolled up per change.
Limit reached → the run pauses and a person decides: raise it, split it, or stop.
One thread per change.
Every update and every escalation is a reply in the same Slack thread — never a new post, never a second place to look.
Bind one automation to one label.
Start attended. Watch the record build. Let the automation earn the rest.