OpsGraph — RepoPulse AIOps

Production-grade AIOps reference. Ingest events from GitHub, OTel logs, and OTel metrics → detect anomalies → group correlated signals into incidents → emit ranked recommendations with explainable evidence → human approval gate → automated workflows with safety guardrails. Operator dashboard included.

Demo

export REPOPULSE_API_SHARED_SECRET="$(openssl rand -hex 16)"
export REPOPULSE_AGENTIC_SHARED_SECRET="$(openssl rand -hex 16)"
./scripts/demo.sh

→ Dashboard at http://127.0.0.1:3000 · API at http://127.0.0.1:8000 (loopback only).

SLO board — availability, error budget, throughput, burn-rate badge	Recommendations inbox — ranked recommendations with approval gate + per-category runbook links

See docs/demo/README.md for the full walkthrough.

What it does

Layer	Module	Tech	TDD'd?
Ingest	repopulse.api.events	FastAPI	✅
Detect	repopulse.anomaly.detector	Modified z-score (Iglewicz & Hoaglin, 1993)	✅
Correlate	repopulse.correlation.engine	Time-window grouping	✅
Recommend	repopulse.recommend.engine	Rule-based + evidence trace	✅
Action gate	repopulse.api.recommendations	Pending → approved/rejected state machine (ADR-004)	✅
Agentic actions	.github/workflows/agentic-*.yml	Kill-switch + scoped tokens (ADR-003)	✅
Operator UI	frontend/	Next.js 15 + Tailwind 4 + Base UI toast	✅

Results

KPIs from backend/scripts/benchmark.py running 4 reproducible scenarios:

KPI	Value
Scenarios run	4
False-positive rate	0%
MTTR (avg, anomaly→trigger)	5.0 s
MTTR (max)	10.0 s
Burn-rate lead time (avg)	0.0 s

Source: docs/superpowers/plans/m6-evidence/benchmark.json · Methodology + per-scenario detail: docs/results-report.md · Re-run command:

cd backend && ./.venv/Scripts/python -m repopulse.scripts.benchmark \
  --scenarios-dir ../scenarios \
  --out ../docs/superpowers/plans/m6-evidence/benchmark.json

Architecture

flowchart LR
  subgraph Sources
    GH[GitHub events]
    OL[OTel logs]
    OM[OTel metrics]
  end
  subgraph Backend["FastAPI (M1–M5)"]
    NORM[normalize] --> ORCH[orchestrator]
    DET[anomaly detector] --> ORCH
    ORCH --> CORR[correlate]
    CORR --> REC[recommend]
    REC --> APP[(approval gate)]
  end
  subgraph UI["Operator dashboard (M4)"]
    SLO[SLO board]
    INC[Incidents]
    INBOX[Inbox]
    HIST[Action history]
  end
  GH --> NORM
  OL --> NORM
  OM --> DET
  REC --> SLO
  REC --> INBOX
  ORCH --> INC
  APP --> HIST

Loading

Per-milestone diagrams: docs/architecture.md · M3 deep-dive: docs/aiops-core.md.

Engineering standards

• TDD across both languages — run cd backend && pytest --co -q and cd frontend && npm test for current counts (see CI for green status).
• Strict typing — mypy strict, TypeScript strict.
• Anti-hallucination — every claim in every milestone handoff has a re-runnable command + captured artifact under docs/superpowers/plans/m<n>-evidence/.
• WCAG 2.2 AA dashboard — live-DOM contrast probe, keyboard verification, semantic HTML before ARIA. See docs/ui-design-system.md.
• Code review per milestone — superpowers:code-reviewer subagent reports filed under each milestone's evidence dir.

Status

Milestone	Tag	Topic
M1	v0.1.0-m1	Foundation, OTel, /healthz
M2	v0.2.0-m2	SLO module, ingest, load generator
M3	v0.3.0-m3	AIOps core (detect + correlate + recommend)
M5	v0.4.0-m5	GitHub agentic workflows (read-only, kill-switch)
M4	v0.5.0-m4	Operator dashboard UI
M6	v1.0.0	Benchmark + portfolio polish
v1.1	v1.1.0	Pipeline API auth, CORS, ingest hardening

Setup + contributing

• docs/SETUP.md — prerequisites + WSL/Docker walkthrough
• docs/CONTRIBUTING.md — workflow + TDD rule
• docs/TROUBLESHOOTING.md — common gotchas

Author

Made by Ibrahim Samad (@Ibrahim4594). Licensed under MIT.

Naming note. Repository is OpsGraph-A; the runtime package + internal branding is RepoPulse. The two will converge in a follow-up rename.