Service control plane
Map CMDB critical services, owners, dependencies, SLAs, 30-day error budgets and multi-window burn-rate SLOs into a single operating model.
// INFINITY SERVICE OPS
Service Operations Engineering for AI-era production systems.
Turn fragile operations into a resilient service control plane: SOE runbooks, multi-burn SLOs, OTel GenAI observability, incident command and AOE automation with MCP 2026-07-28 & DORA/NIS2 guardrails.
SOE · Service Operations EngineeringAOE · AI Operations EngineeringMCP-ready automation
∞
SLOMCPPIRLMT24/7Ops-ready patternsincident, change, DORA/NIS2
SLOReliability targets14.4x / 6x multi-burn + ownership
MCPAI tool operationsOAuth 2.1, ToolAnnotations, HITL
30dCutover roadmapfrom discovery to cadence
// FREE DEV TOOL · #OPSFORGE
OpsForge Service Blueprint & Config Generator
Interactive readiness canvas and configuration generator for MCP-era service operations. Calculate Google SRE multi-window SLO burn alerts, export CNCF OpenTelemetry gen_ai.* pipelines, enforce MCP OAuth 2.1 tool gates, and generate EU DORA / NIS2 incident command runbooks instantly—no sign-up.
01
Inventory CMDB service ownership, integrations, data flows and failure modes.
02
Score current SOE maturity across multi-burn SLOs, incident, change and OTel coverage.
03
Select AOE / MCP automation candidates with low blast radius and high operator value.
04
Ship a 30-day action plan with multi-window alerts, runbooks, HITL controls and DORA/NIS2 cadence.
30-DAY ERROR BUDGET43.2m0.10% allowed failure ratio
14.4x FAST PAGE (1h & 5m)> 1.440% err2% budget burned in 1h (50h cap)
6.0x SUSTAINED PAGE (6h & 30m)> 0.600% err5% budget burned in 6h (5d cap)
STATUTORY / SLA CLOCK<= 4h Initial / <= 1mo FinalDORA escalation window
| Alert Tier | Burn Rate | Long + Short Window | Trigger Error Rate | 30d Budget Consumed | Operator Action |
|---|---|---|---|---|---|
| PAGE (SEV-1) | 14.4x | 1h + 5m | > 1.440% | 2% (50h (~2.08d)) | Wake on-call IC immediately; freeze deploys & verify rollback manifest |
| PAGE (SEV-2) | 6x | 6h + 30m | > 0.600% | 5% (120h (5d)) | Page primary engineer; run read-only MCP diagnostics & canary check |
| TICKET (SEV-3) | 3x | 1d + 2h | > 0.300% | 10% (240h (10d)) | Open next-business-day ITSM ticket; inspect OTel trace tail-samples |
| TICKET (SEV-4) | 1x | 3d + 6h | > 0.1000% | 10% (720h (30d)) | Track capacity/prompt drift in weekly SOE reliability review |
# OpsForge Generated — Google SRE Workbook Multi-Window Multi-Burn-Rate SLO Rules
# Service: payments-gateway | Target: 99.9% (30d window) | Error Budget: 0.10% (43.2m)
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
name: payments-gateway-slo-multiburn
labels:
service: payments-gateway
slo.target: "99.9"
opsforge.schema: "sre-workbook-ch5-2026"
spec:
groups:
- name: payments-gateway.slo.burn_rate
interval: 30s
rules:
# Tier 1: 14.4x burn over 1h AND 5m -> 2% of 30d error budget consumed (50h to exhaustion)
- alert: SLOErrorBudgetBurnCriticalPage
expr: |
(
job:slo_http_errors:ratio_rate1h{service="payments-gateway"} > (14.4 * 0.001)
and
job:slo_http_errors:ratio_rate5m{service="payments-gateway"} > (14.4 * 0.001)
)
for: 2m
labels:
severity: critical
route: pager
burn_rate: "14.4x"
annotations:
summary: "payments-gateway: 14.4x SLO burn (1.440% error rate across 1h & 5m)"
description: "Consuming 2% of 30-day error budget per hour. Trigger SEV-1 IC & inspect last release manifest."
# Tier 2: 6.0x burn over 6h AND 30m -> 5% of 30d error budget consumed (5d to exhaustion)
- alert: SLOErrorBudgetBurnHighPage
expr: |
(
job:slo_http_errors:ratio_rate6h{service="payments-gateway"} > (6 * 0.001)
and
job:slo_http_errors:ratio_rate30m{service="payments-gateway"} > (6 * 0.001)
)
for: 5m
labels:
severity: high
route: pager
burn_rate: "6x"
annotations:
summary: "payments-gateway: 6x sustained SLO burn (0.600% error rate across 6h & 30m)"
# Tier 3 & 4: Ticket-level erosion (3x over 1d/2h = 10% budget; 1x over 3d/6h = 10% budget)
- alert: SLOErrorBudgetSlowErosionTicket
expr: |
(
job:slo_http_errors:ratio_rate1d{service="payments-gateway"} > (3 * 0.001)
and
job:slo_http_errors:ratio_rate2h{service="payments-gateway"} > (3 * 0.001)
)
or
(
job:slo_http_errors:ratio_rate3d{service="payments-gateway"} > (1 * 0.001)
and
job:slo_http_errors:ratio_rate6h{service="payments-gateway"} > (1 * 0.001)
)
for: 15m
labels:
severity: warning
route: itsm-ticket
annotations:
summary: "payments-gateway: Slow error-budget erosion (>10% 30d budget projected)"
# Latency SLI Gate (p95 <= 500ms)
- alert: SLOLatencyP95Breach
expr: |
histogram_quantile(0.95, sum by (le) (rate(http_server_request_duration_seconds_bucket{service="payments-gateway"}[5m]))) > 0.500
for: 5m
labels:
severity: high
sli: latency_p95// SOE
Service Operations Engineering
SOE creates the operating system for your services: who owns what, how reliability is measured, how incidents are handled and how knowledge becomes repeatable execution.
Incident command system
Define severity levels, commander roles, stakeholder comms, post-incident review loops and EU DORA (<=4h) / NIS2 (<=24h/72h) escalation clocks.
Runbook engineering
Convert tribal knowledge into tested runbooks with whole-manifest rollbacks (code, prompt, model, guardrail), evidence capture and automation gates.
Observability topology
Design CNCF OpenTelemetry OTLP 1.x logs, metrics, traces and gen_ai.* semantic conventions aligned to real user journeys instead of noisy hosts.
// AOE
AI Operations Engineering
AOE adds AI assistance without losing control. Agents can read, reason and recommend first; controlled execution comes only after permissions, evals, telemetry and rollback are proven.
MCP-aware ops automation
Expose approved operational actions over stateless Streamable HTTP (MCP 2026-07-28) with OAuth 2.1 + PKCE, RFC 8707 audience binding, ToolAnnotations and HITL gates.
AI support copilot design
Build support assistants that retrieve versioned runbooks, correlate OTel traces, summarise incidents, draft updates and recommend verified next checks.
Agent evaluation harness
Test AI operations workflows against golden tickets, destructiveHint traps, prompt-injection payloads, latency budgets and recovery scenarios.
Automation adoption path
Move from readOnlyHint insight to guided actions, then controlled scoped execution once OAuth scopes, evals, approvals and rollback are mature.
// DELIVERY MODEL
From noisy support to reliable service cadence
Discover
Stabilise
Instrument
Automate
Review
// FAQ
Service Ops FAQ
Who is Infinity Service Ops for?+
Founders, platform teams, IT leaders and support organisations that need calmer production operations, clearer ownership, better incident response and practical AI automation.
Does this replace my current ITSM tooling?+
No. The first goal is to improve the operating model around your tools. Automation and MCP integrations can then connect Jira, ServiceNow, GitHub, Slack, Teams or custom systems safely.
What is delivered in an SOE engagement?+
A service map, SLO catalogue, incident model, runbook set, observability recommendations, escalation matrix and practical improvement backlog.
What is delivered in an AOE engagement?+
An AI operations automation design: tool permissions, prompt and retrieval architecture, eval cases, guardrails, telemetry, rollout stages and acceptance criteria.
How does this connect with MCP-related tools?+
Service Ops uses MCP 2026-07-28 Stateless Streamable HTTP patterns, OAuth 2.1 + PKCE (RFC 9728 & RFC 8707), and ToolAnnotations to make operational tools discoverable, permissioned and auditable without bypassing service controls.
How does OpsForge calculate multi-window, multi-burn-rate SLO alerts?+
OpsForge implements Chapter 5 of the Google SRE Workbook across a 30-day compliance window: 14.4x burn over 1h and 5m pages immediately (consuming 2% of monthly error budget), 6x burn over 6h and 30m pages on sustained degradation (5% budget), and 3x (1d/2h) plus 1x (3d/6h) open next-day tickets at 10% budget erosion.
Which OpenTelemetry standards are used for AI and service observability?+
Blueprints target CNCF Graduated OpenTelemetry over OTLP 1.x (ports 4317 gRPC / 4318 HTTP) alongside OpenTelemetry GenAI Semantic Conventions (gen_ai.operation.name, gen_ai.provider.name, gen_ai.client.token.usage, and gen_ai.client.operation.duration), keeping raw prompt/completion capture opt-in and PII-redacted.
How are MCP ToolAnnotations and human-in-the-loop gates enforced?+
Under the Model Context Protocol specification, readOnlyHint, destructiveHint, idempotentHint and openWorldHint are untrusted client hints rather than cryptographic guarantees. OpsForge pairs them with mandatory server-side OAuth 2.1 scopes, RFC 8707 audience checks (zero token passthrough), SEP-2243 Mcp-Method/Mcp-Name gateway headers, and human-in-the-loop Incident Commander approval before any mutating action executes.
Does OpsForge cover EU DORA and NIS2 operational resilience reporting?+
Yes. OpsForge generates incident command runbooks aligned with EU Regulation 2022/2554 (DORA, fully applicable since 17 January 2025: initial notification within 4h of major ICT classification and <=24h from awareness, intermediate updates, and 1-month final report) as well as EU Directive 2022/2555 (NIS2: 24h early warning, 72h incident notification, and 1-month final report).
// NEXT STEP
Build the calm, observable, AI-assisted ops layer.
Start with the free blueprint, then turn the findings into an SOE/AOE roadmap your team can ship.