Skip to content

Use cases

Oncall cover for repository work

Issues do not arrive between nine and five. The agent handles the routine ones around the clock and wakes you only for the classes of problem you said should wake you.

Audit / task_01J2XLIVE
Issue assignedscope captured
09:41:02
Plan accepted4 files · 2 tests
09:41:11
Patch created+84 −16
09:43:48
Checks passed12 / 12
09:45:03
PR readyreceipt $1.42
09:45:07
sha256: 1d8a…9b2f · verified
Audited end to endGitHub-native workflowExisting checks stay in controlCost visible per task

Escalate, don't act

The default for anything sensitive is to stop and notify, not to proceed. Security reports, licence questions, hostile threads and anything your policy marks as human-merge are escalated to you rather than answered. Everything else — the triage, the follow-up question, the bump PR, the red check — is handled while you sleep, and is waiting in the audit log with its reasoning when you wake up.

A defined escalation set

Security, licence and tone signals are escalation classes, not judgement calls made fresh each time.

Notified where you already are

Escalations reach you by email today, with Slack routing on the roadmap rather than in the copy as though it shipped.

Morning is a read, not an archaeology dig

Everything the agent did overnight is in the audit log with its input, its output and its reason.

Go deeper

The escalations doc lists every class that pages a human and how to add your own.

Read the escalations doc

Put a fleet of AI developers to work

Bring your own keys and your own boxes, drop a .maintainer.yml, and let the loops run — audited end to end. Access is invite-only.