Support · Converge

How AI support works here

the agent owns the procedure — you own the decisions

1 · DETECT

Failed step, alert, or detection — captured with full diagnostics at the moment it happens.

2 · DIAGNOSE

The agent reads the bundle, cites evidence, and matches the failure signature against the fleet-wide known-issue base.

3 · FIX & RETRY

Heals the classes it can — stale locks, throttles, missing certs — with bounded retries. Never thrashes.

4 · ESCALATE

When it must ask, it asks one precise question — a decision or a secret.

5 · LEARN

Every failure ships a structured signature to the vendor — the engine improves, so the next site never hits it.

Cases

Case	Summary	Environment	Sev	Status
CASE-1042	Search OOMKilled after 0.1.90 ingest change auto-opened from INC-241 · bundle + agent diagnosis attached · known-issue match: fix staged for 0.1.92	kestrel-bank/prod	2	awaiting vendor
CASE-1038	Vendor analysis of run-8842 verify warnings vendor requested the migration log — export awaiting your approval	meridian-health/prod	4	awaiting your approval
CASE-1031	Air-gap bundle sync guidance answered — runbook RB-024 updated and shipped with 0.1.91 docs	cobalt-mining/prod	4	closed
CASE-1027	West Europe quota pre-check false negative became engine fix: per-family vCPU preflight in 0.1.91 — closed fleet-wide	helios-energy/prod	3	closed · engine fix

Bundle exchange — CASE-1038

needs your approval

The vendor asked for the migration log from run-8842. Nothing leaves your boundary until you approve exactly what's inside:

bundle-case-1038.tgz · 4.2 MB · signed
  ✓ migration log (run-8842, steps 4–6)
  ✓ helm release history (values redacted: 14 secrets masked)
  ✓ k8s events, ns=platform, 03:00–15:00
  ✗ excluded: tenant data, PII, credentials, TLS keys

Air-gapped sites (cobalt-mining) export the same bundle to file and carry it out on their own terms.

Agent-assisted operations

install by talking to an engineer

The same agent that triages incidents drives installs and upgrades through 6 typed MCP tools — it validates prerequisites before step one, gathers config progressively, and asks only what the target actually requires.

Last assisted install

sable-insurance / prod · AWS eu-central-1 · 47 min, 2 questions asked

Self-healed on the way

TLS cert issued · stale tf lock cleared · API throttle retried

Guide pages read

Open Copilot

Entitlement & SLA

Severity	First response	Coverage	30d actual
sev-1 production down	15 min	24×7	no sev-1
sev-2 degraded	1 h	24×7	28 min avg
sev-3 impaired	4 h	business hours	1.9 h avg
sev-4 question	1 business day	business hours	3.1 h avg

Enterprise plan · renewal 2027-02-01 · named TAM: Jide Okafor · quarterly architecture review included · sev-1/2 auto-open a vendor case with the bundle already attached (after your approval policy).

Known-issue base

fleet-wide signatures · synced with the release

Signature	Seen at	Resolution
search OOMKilled · ingest ≥ 2× batch KI-2026-018 · matches INC-241	3 sites	fix staged → 0.1.92
DNS-01 throttle during cert renewal KI-2026-015	5 sites	self-heal rule shipped 0.1.90
per-family vCPU quota false pass KI-2026-011 · from CASE-1027	2 sites	preflight check shipped 0.1.91
AKS destroy→recreate name race KI-2026-006	1 site	recovery logic shipped 0.1.68

Incidents check against fleet-wide failure signatures at diagnosis time. Failures you hit feed the base in return — as structured signal, never raw data.