NeuroOps is an autonomous AI SRE engine that detects, diagnoses, and remediates Kubernetes incidents end-to-end — with zero human intervention for high-confidence scenarios.

---

📊 Results

Metric	Result
🎯 Chaos Incidents Resolved	15 / 15 — 100%
⚡ Average MTTR	< 4 minutes
🏆 DORA Tier	Elite Performer
🎪 False Positive Rate	0%
💰 Cost vs Manual On-Call	> 1,600× cheaper
🤖 Autonomous Resolution Rate	50–100% (confidence-gated)

---

🧠 How It Works

flowchart TD
    A([🔥 Fault Injected]) --> B

    subgraph DETECT["🔍 DETECT"]
        B[Prometheus Scrapes\nGolden Signals] --> C[Isolation Forest\nAnomaly Scoring]
        C --> D{Anomaly?}
        D -- No --> B
        D -- Yes --> E[Alert Dedup\n+ Severity Triage]
    end

    E --> F

    subgraph DIAGNOSE["🧬 DIAGNOSE — LangGraph Fan-Out"]
        F[Supervisor Init\nIncident ID + OTel Trace] --> G1 & G2 & G3 & G4
        G1[🔎 Detective\nMetric Correlation]
        G2[🗺️ Topologist\nTrace Analysis]
        G3[📜 Historian\nDeploy Timeline]
        G4[📋 Log Analyser\nPod Logs]
        G1 & G2 & G3 & G4 --> H[🧠 Supervisor\nRoot Cause + Confidence]
    end

    H -- "< 0.55" --> I[👤 Escalate\nto Human]
    H -- ">= 0.65" --> J

    subgraph REMEDIATE["🔧 REMEDIATE"]
        J{Action?} -- "restart / scale" --> K[⚡ Auto-Execute]
        J -- "rollback / PR" --> L[🔔 Slack Approval\nThen Execute]
        K & L --> M[✅ Verify Resolution]
    end

    M --> N[📄 Post-Mortem\nStored]

    style DETECT fill:#1a1a2e,stroke:#00D9FF,color:#fff
    style DIAGNOSE fill:#1a1a2e,stroke:#FF6B35,color:#fff
    style REMEDIATE fill:#1a1a2e,stroke:#00C853,color:#fff

Loading

---

🏗️ Architecture

graph TB
    subgraph UI["🌐 Web UI :3000"]
        WEB[Orbital Command Hub\nReal-time Dashboard]
    end

    subgraph CORE["⚙️ Core Services"]
        DET["🔍 Detector :8001"]
        AGT["🧠 Agent :8002"]
        REM["🔧 Remediator :8003"]
    end

    subgraph OBS["📡 Observability"]
        PROM[Prometheus]
        JAEGER[Jaeger]
        GRAFANA[Grafana]
        OTEL[OTel Collector]
    end

    subgraph K8S["☸️ Kubernetes"]
        WORKLOADS[Target Workloads]
        CHAOS[LitmusChaos]
    end

    WEB --> AGT & DET
    DET --> PROM & OTEL
    AGT --> PROM & JAEGER & OTEL
    REM --> WORKLOADS & OTEL
    CHAOS --> WORKLOADS
    WORKLOADS --> PROM
    OTEL --> JAEGER
    PROM & JAEGER --> GRAFANA

    style UI fill:#0d1117,stroke:#00D9FF,color:#00D9FF
    style CORE fill:#0d1117,stroke:#FF6B35,color:#FF6B35
    style OBS fill:#0d1117,stroke:#7C3AED,color:#7C3AED
    style K8S fill:#0d1117,stroke:#326CE5,color:#326CE5

Loading

---

🚀 Quickstart

Prerequisites: Python 3.11+, Docker, Kubernetes cluster (Minikube / kind / EKS)

git clone https://github.com/Tayab-Ahamed/neuroops.git
cd neuroops
cp .env.example .env      # fill in your API keys

Docker Compose — full stack

docker compose up --build

Services: :8001 Detector · :8002 Agent · :8003 Remediator
Open web-ui/index.html for the live dashboard.

Run services individually

# Detector
cd detector && pip install -r requirements.txt && uvicorn server:app --port 8001

# Agent
cd agent && pip install -r requirements.txt && uvicorn main:app --port 8002

# Remediator
cd remediator && pip install -r requirements.txt && uvicorn server:app --port 8003

---

🔑 Environment Variables

# LLM
ANTHROPIC_API_KEY=sk-ant-...          # Required — RCA agents
OPENAI_API_KEY=sk-...                 # Optional fallback

# Observability
PROMETHEUS_URL=http://localhost:9090
JAEGER_QUERY_URL=http://localhost:16686
OTEL_COLLECTOR_ENDPOINT=http://localhost:4317

# Kubernetes
KUBECONFIG=~/.kube/config
TARGET_NAMESPACE=neuroops-demo

# GitHub
GITHUB_TOKEN=ghp_...                  # Historian agent + PR actions
GITHUB_REPO=your-username/repo

# Tuning
CONFIDENCE_THRESHOLD=0.65             # Below → human escalation
AUTONOMOUS_CONFIDENCE_THRESHOLD=0.65  # Actions above this run autonomously
ANOMALY_CONTAMINATION=0.05

# ChatOps
SLACK_WEBHOOK_URL=https://hooks.slack.com/...   # Optional

---

📡 API Reference

Detector — :8001

Method	Endpoint	Description
GET	/health	Health + anomaly model status
GET	/alerts	Active alerts
GET	/metrics	Prometheus scrape endpoint
POST	/baseline/train	Trigger baseline training

Agent — :8002

Method	Endpoint	Description
POST	/investigate	Trigger RCA for an alert
GET	/incidents	All persisted incidents
GET	/incidents/{id}	Single incident + full RCA trace
GET	/incidents/{id}/similar	Top-K similar incidents (RAG)
GET	/analytics/mttr	p50/p95/p99 MTTR per service
GET	/analytics/sla	SLA breach + autonomous resolution rate
GET	/analytics/cost	LLM token + USD cost tracking

Remediator — :8003

Method	Endpoint	Description
POST	/remediate	Execute remediation action
GET	/health	Health + action count
GET	/metrics	Prometheus scrape endpoint

---

📁 Project Structure

neuroops/
├── detector/              # Anomaly detection — Isolation Forest + Ridge Regression
├── agent/                 # LangGraph RCA — Detective, Topologist, Historian, Supervisor
├── remediator/            # Actions — restart, rollback, scale, patch, PR + approval gate
├── observability/         # OTel collector config, Grafana dashboards, CLI replay tool
├── benchmarks/            # Chaos benchmark runner + report generator
├── web-ui/                # Orbital Command Hub (real-time HTML dashboard)
├── cluster/               # Kubernetes manifests + LitmusChaos experiments
├── docker-compose.yml
├── Makefile
└── pyproject.toml

---

📜 License

MIT — see LICENSE

---

_{Built by Tayab Ahamed}