Glossary
Agent. The Seylo software you install in each cluster. It connects out to Seylo over one connection and does what Seylo asks within the RBAC you granted. Not to be confused with Nightshift or the coding agent.
Anchor alert. The alert an incident’s Nightshift run starts from and is kept on: the root’s alert when the walk named a root, else the earliest open member.
Approval. A request to change something, waiting for an approver. One
approval is enough, and approving your own request is allowed. See
approvals.
Architecture file. A Backstage catalog-info.yaml or a Seylo-format file
that tells the service map
what the cluster cannot show: functions, managed databases and queues, SaaS,
asynchronous hops.
Automation. The page, under Operate, that lists what Seylo may change on its own and everything it did. See Automation.
Blast radius. The services that depend on an incident’s failure: affected when they alert or their calls fail, at risk when they depend on it and are quiet.
Coding agent. An AI coding assistant that works through Seylo: one you run and connect over MCP, or the hosted coding agent Seylo runs for you. It has its own model configuration, separate from Nightshift’s.
Core runbook. One of Seylo’s 45 built-in runbooks, the same for every organisation, each in the order mitigate, diagnose, fix, prevent. Your skills come before them. See core runbooks.
Coverage. Which workloads send traces, and which serve traffic and do not. See APM coverage.
Deploy token. A credential for one cluster and namespace, with chosen capabilities, that a coding agent uses. See tokens.
Destructive tier. The agent’s RBAC tier that may delete claims, volumes,
StorageClasses, namespaces and CRDs (rbac.write.destructive.enabled). Off by
default. See RBAC posture.
Door. How a fix reaches a workload: in the cluster, as a pull request to the Git repository its owner deploys from, or as a task for the coding agent. See the fix on the root.
Earned autonomy. A class of fix that may run on its own after it was
applied and verified enough times in your organisation, while an admin’s
envelope is on. See fixes, class by class.
Evidence. What Nightshift read, numbered and redacted. Every finding must cite it by number. See Nightshift.
Fix. A change Nightshift proposes with its evidence, the exact diff, how it will be verified and how it would be undone. See Nightshift’s fixes.
Generated fix. A change Nightshift’s model proposes when no fixed class of fix applies, accepted only by a five-check guard. See generated fixes.
Grant. A permission an admin requests
and an approver signs, which lets Seylo make named changes on its own inside
a namespace, window, rate limit and end date. Listed on
Automation. Called an autoscale policy or a standing
permission before 0.4.9.
Health signal. What an exporter says about a pool, connection limit, lag, backlog, retry or breaker, judged failing, degraded, healthy or unknown. See health signals.
IdP. Identity provider: the system your people sign in with, such as Entra ID, Okta or Google Workspace, connected through single sign-on.
Incident. Two or more related alerts that the service map connects, within minutes of each other, treated as one failure. See incidents.
Investigation (run). One bounded pass of Nightshift over an alert or an incident: reads, its own queries, findings and proposals. See investigate and fix.
Lesson. One entry in Nightshift’s memory: the symptom, the cause, the fix and whether it held.
Managed savings. Seylo applying a cluster’s savings levers on its own,
inside a grant an admin signs for 90 days, with each change checked and put
back if it does not hold. Off by default. See
managed savings.
Mitigation. A step that limits the damage before the cause is fixed: a capacity buffer, a rate limit, a maintenance page, scaling down another workload. Always a person’s approval. See mitigations.
Nightshift. Seylo’s AI SRE, which investigates alerts and incidents and proposes fixes. It has its own group in the portal’s sidebar. See Nightshift.
Organisation. Your company’s space in Seylo: its people, clusters, data and settings. See create your organisation.
Playbook. What Seylo would do about a kind of alert, and whether it may do it on its own, shown under every investigation. See fix.
Posture. Two things: the agent’s RBAC posture (read-only, deploy,
admin; see RBAC posture), and the Posture
page under Admin, which checks your workloads against policies and runs the
production-readiness campaign.
A finding on Seylo’s own node-exporter, for the host network and read-only host
mounts it needs, is listed as exempt: Seylo’s own with the reason and left
out of the score. See
on the Posture page.
Reconciler. Argo CD, Flux or Helm: a tool that keeps a resource as its source says. Seylo refuses to change a reconciler-owned resource until you acknowledge the owner by name, and never changes one unattended.
Root. The deepest failing service in an incident whose own dependencies are healthy: where the failure started. See the root-cause walk.
Service map. What each workload calls and what calls it, each edge with its source and confidence. See service map.
Silence. A rule that holds back notifications for matching alerts, with a reason and an end at most 90 days away. See alerts.
Skill. One of your organisation’s runbooks, as a SKILL.md file, that
Nightshift can load and cite. See skills.
Snooze. Quiets one alert’s notifications for up to 7 days. See alerts.
Verified fix. A fix whose check passed for its whole window after it was applied. A fix that does not verify is undone where it has an undo.
Watched namespaces. The namespaces an agent manages (watchNamespaces).
On a namespace-scoped install, the only namespaces it can see at all. See
RBAC posture.