An AI-driven platform that predicts infrastructure failures, identifies root causes, and automatically executes self-healing actions on Kubernetes clusters.
The Fast Way:
./start-dev.ps1For detailed instructions, manual setup, and Docker options, see QUICKSTART.md.
Run the end-to-end test suite:
python scripts/test_scenario.pySee TESTING.md for more details.
AntiGravity DevOps Assistant is an intelligent platform that combines machine learning, rule-based analysis, and automated remediation to maintain Kubernetes workloads proactively. It reduces manual SRE work by predicting failures before they occur and executing safe self-healing actions autonomously.
- Predictive Monitoring: ML models predict incidents 0-5 minutes before they occur
- Rule-Based RCA: Automated root cause analysis using decision trees and pattern matching
- Self-Healing: Automated Kubernetes actions with built-in safety constraints
- Real-Time Dashboard: Beautiful brown-themed UI with live updates
- 100% Open Source: No vendor lock-in, runs entirely on free tools
graph TD
subgraph "Infrastructure"
K8s[Kubernetes Cluster]
Prom[Prometheus]
OpenS[OpenSearch]
Postgres[(PostgreSQL)]
Redis[(Redis)]
end
subgraph "Core Services"
Obs[Observability Service]
ML[ML Prediction Service]
Analysis[Analysis Service]
Heal[Self-Healing Engine]
Gateway[API Gateway]
end
Fe[Frontend Dashboard]
%% Data Flow
K8s -->|Metrics| Prom
K8s -->|Logs| OpenS
Prom -->|Scrape| Obs
OpenS -->|Query| Obs
Obs -->|Features| ML
ML -->|Prediction| Analysis
Obs -->|Anomalies| Analysis
Analysis -->|Action Plan| Heal
Heal -->|Execute| K8s
%% API Flow
Fe -->|REST/WS| Gateway
Gateway --> Obs
Gateway --> ML
Gateway --> Analysis
Gateway --> Heal
%% Storage
Obs --> Postgres
Analysis --> Postgres
Heal --> Postgres
Obs --> Redis
ML --> Redis
- Observability Service: Collects metrics (Prometheus) and logs (OpenSearch)
- ML Prediction Service: Runs RandomForest & LSTM models for failure prediction
- Analysis Service: Rule-based RCA and remediation planning
- Self-Healing Engine: Executes safe Kubernetes actions
- API Gateway: Single entry point with WebSocket support
- Environment Variables
Copy the example configuration to create your local secrets file:
Edit
cp .env.example .env
.envto set your own passwords and ports. Use the default values for a quick start.
Backend: FastAPI, Python 3.11, PyTorch, scikit-learn, SQLAlchemy, Celery
Frontend: Next.js 14, TypeScript, TailwindCSS, shadcn/ui, Recharts
Infrastructure: K3d (Kubernetes), Prometheus, OpenSearch, PostgreSQL, Redis
Monitoring: Grafana, Fluent Bit
devops-assistant/
├── services/
│ ├── observability/ # Metrics + logs collection
│ ├── ml-prediction/ # ML models and inference
│ ├── analysis/ # Rule-based RCA
│ ├── self-healing/ # Kubernetes automation
│ └── api-gateway/ # API routing + WebSocket
├── frontend/ # Next.js dashboard
├── k8s/ # Kubernetes manifests
├── scripts/ # Setup and deployment scripts
├── migrations/ # Database migrations
└── docker-compose.yml # Local development
- Home: Cluster health overview with real-time metrics
- Predictions: ML prediction timeline and confidence scores
- Logs: Advanced search and filtering
- Actions: Timeline of executed self-healing actions
- Explainability: Root cause analysis details
# Get current cluster metrics
curl http://localhost:8000/api/v1/metrics/current
# Get predictions
curl http://localhost:8000/api/v1/predictions
# Get executed actions
curl http://localhost:8000/api/v1/actions
# WebSocket for real-time updates
wscat -c ws://localhost:8000/ws/events- Namespace Protection: Never operates on
kube-system - Rate Limiting: Max 3 actions per hour
- Confidence Threshold: Auto-execution requires ≥75% confidence
- Audit Logging: All actions logged with timestamp and reasoning
# Run all tests
make test
# Run specific service tests
cd services/ml-prediction && pytest
# Load testing
make load-test- Prediction Accuracy: ≥75%
- Prediction Latency: <1 second
- Correct Remediation: ≥80%
- Dashboard Load Time: <1 second
- Autonomous Resolution: ≥70% of incidents
- K3d cluster with sample apps
- ML-based failure prediction
- Rule-based RCA
- Automated self-healing
- Real-time dashboard
- Reinforcement learning for remediation
- Multi-cluster support
- LLM-driven auto-scaling
- Discord/Slack integration
- Vector database for log anomalies
MIT License - see LICENSE
Built with love using amazing open-source tools: Kubernetes, FastAPI, Next.js, PyTorch, and many more.
Made by a team of AI-curious engineers