Maestro

Runs now report they are alive while they work, and anything that stops reporting is closed out rather than sitting as running forever. Schedules claim their slot in a single step, so a restart, a second copy of the service, or a failure part-way through a tick cannot produce the same run twice. Long downtime replays at most one run per schedule instead of a burst.

A systematic read of the runtime, the API, the integrations and the database turned up twenty-four issues, several of them security-relevant. The most serious were fixable straight away and were: credentials can no longer be read out without authorisation, the general-purpose web step can no longer be pointed at internal addresses, and outgoing mail headers are sanitised. Two thirds of the list is now written and verified, and none of it has shipped yet.

How an integration, an agent, an automation and a schedule are authored is now written down in a form both a person and a machine can read, and there is a single definition of what makes an automation valid — checked in one place instead of three. This is the groundwork for describing an automation in plain English and having it drafted for you.

A canvas replaces the old edit-the-seed-data workflow. Steps get dragged on, connections drawn between them, and each step configured where it sits. Changing what an automation does no longer means touching the database.

Maestro runs end to end: it finds leads, writes to them from the operator's own inbox, reads what comes back, and keeps its own record of every contact and interaction. Everything a customer would see waits for approval first. Three closed-beta partners are running it on their own machines.