Skip to content

Alerts and incidents

Alert rules for spend caps, failing runs, stuck leases and a stale index, routed to email, Slack and Teams, with one notice per incident.

Updated View as MarkdownOpen in ChatGPTOpen in Claude

A rule names a condition and where its notice goes. When the condition appears, an incident opens and the notice is sent once. Seeing the same condition again, a replayed event or two checks at the same moment opens nothing new.

overlane observe alerts set --kind failing_run --email ops@example.com --slack ops
overlane observe alerts set --kind spend_cap --percent 80 --email ops@example.com
overlane observe alerts
overlane observe alerts evaluate

The four conditions

Kind The condition Threshold
spend_cap A budget or spending cap has been used up to the percent --percent, default 100
failing_run A run ended in failure in the last window --minutes, default 1440
stuck_lease A task lease is held with no run activity, or a running run has gone silent --minutes, default 30
index_stale The code index is marked stale, or an index job has waited too long --minutes, default 1440

Rules are checked on a schedule, when a run ends in failure and when cost arrives, and on demand with observe alerts evaluate.

Once per incident

An incident is one condition: a spend line in a period, one failed run, one stuck lease. The database allows one open incident per rule and condition, so the notice goes out once however often the condition is seen. A notice that could not be delivered stays on the incident and is retried on the next check; a delivered one is not sent again. When the condition clears, the incident resolves and a separate resolved notice goes out, once. A later failure is a new incident.

The notice

A notice is structured text, one key: value per line, ending with the command to run:

incident: inc_...
kind: failing_run
state: opened
subject: run run-12 by codex failed
detail: project prj_1, ended 2026-10-10T10:00:00.000Z
next: overlane observe logs --run run-12

Email goes through the Email Service the invitations use. Slack and Teams notices are posted through the team’s connected Slack or Teams connection; without one, the incident records the failed delivery and nothing else is affected. A rule stores a channel name or an address, never a webhook URL or a token.

Command reference: observe alerts, observe alerts set, observe alerts remove and observe alerts evaluate.

Navigation

Type to search…

↑↓ navigate↵ selectEsc close

Overlane Agent

Ask about the docs. Answers are quoted from the pages, with links. This panel cannot change anything.

    Questions are answered in your browser from the docs index. Nothing you type here is stored or sent anywhere.