Self-Hosted Installation Overview

CVAT offers flexible deployment options to suit different infrastructure requirements, from local development to enterprise-scale production environments.

​

Deployment Options

​

Docker Compose

Best for: Small to medium teams, development, testing, and quick deployments Pros:

• Simple setup with minimal configuration
• Easy to run on a single machine
• Quick deployment (5-10 minutes)
• Ideal for evaluation and proof-of-concept
• Lower resource requirements

Cons:

• Limited scalability
• No built-in high availability
• Manual scaling of components

Requirements:

• Docker Engine 20.10+
• Docker Compose 2.x
• 4GB RAM minimum (8GB recommended)
• 2 CPU cores minimum (4 recommended)
• 50GB disk space

Learn more →

​

Kubernetes with Helm

Best for: Production environments, enterprise deployments, auto-scaling needs Pros:

• Horizontal scaling and load balancing
• High availability and fault tolerance
• Resource management and optimization
• Rolling updates with zero downtime
• Cloud-native deployment

Cons:

• More complex setup and maintenance
• Requires Kubernetes expertise
• Higher infrastructure overhead

Requirements:

• Kubernetes 1.23+
• Helm 3.x
• kubectl configured
• Persistent storage provider
• Ingress controller (optional but recommended)

Learn more →

​

AWS Deployment

Best for: Cloud deployments with GPU support for auto-annotation Pros:

• Elastic infrastructure
• GPU instance support (P3 instances)
• Integration with AWS services (S3, RDS, EFS)
• Managed database options

Cons:

• Cloud costs
• AWS-specific configuration

Learn more →

​

Architecture Overview

CVAT’s self-hosted architecture consists of several key components:

​

Core Services

CVAT Server (cvat/server)

• Django-based API server
• Handles authentication, authorization, and business logic
• Runs on port 8080
• Manages task creation, annotation, and exports

CVAT UI (cvat/ui)

• React-based frontend application
• Serves the web interface on port 8000
• Provides annotation tools and project management UI

Reverse Proxy (Traefik)

• Routes traffic to appropriate services
• Handles SSL/TLS termination
• Load balancing and service discovery
• Default port: 8080 (HTTP), 443 (HTTPS)

​

Data Layer

PostgreSQL (postgres:15-alpine)

• Primary relational database
• Stores users, projects, tasks, and metadata
• Port: 5432

Redis (redis:7.2.11-alpine)

• In-memory cache for session management
• Job queue for background tasks
• Port: 6379

Kvrocks (apache/kvrocks:2.12.1)

• Redis-compatible on-disk storage
• Caches media chunks and temporary data
• Port: 6666

ClickHouse (clickhouse/clickhouse-server:23.11-alpine)

• Analytics and event logging
• Stores user actions and system events
• Port: 8123

​

Worker Services

CVAT uses multiple worker processes to handle background tasks asynchronously:

• cvat_worker_import: Handles dataset and annotation imports (2 processes)
• cvat_worker_export: Processes annotation exports (2 processes)
• cvat_worker_annotation: Manages annotation operations (1 process)
• cvat_worker_webhooks: Sends webhook notifications (1 process)
• cvat_worker_quality_reports: Generates quality analytics (1 process)
• cvat_worker_chunks: Processes video/image chunks (2 processes)
• cvat_worker_consensus: Handles consensus annotation calculations (1 process)
• cvat_worker_utils: Notifications and cleaning tasks (1 process)

​

Analytics Stack

Vector (timberio/vector:0.26.0-alpine)

• Log aggregation and forwarding
• Sends events to ClickHouse
• Port: 8282

Grafana (grafana/grafana-oss:10.1.2)

• Analytics dashboard
• Visualizes user activity and system metrics
• Accessible at /analytics

​

Security & Policy

Open Policy Agent (openpolicyagent/opa:1.12.2)

• Policy-based access control
• Dynamic permission evaluation
• Port: 8181

​

Data Flow

1. User Request: Browser → Traefik → CVAT UI/Server
2. Authentication: Server → PostgreSQL (user validation)
3. Session Management: Server → Redis (session cache)
4. Data Processing: Server → Workers (background jobs)
5. Media Storage: Server → Shared volumes → Kvrocks cache
6. Analytics: Server → Vector → ClickHouse → Grafana
7. Authorization: Server → OPA → Decision

​

Storage Requirements

​

Docker Volumes

• cvat_db: PostgreSQL data (~1-5GB depending on usage)
• cvat_data: Media files and datasets (varies greatly, plan for 100GB+)
• cvat_keys: Authentication keys and secrets (~1MB)
• cvat_logs: Application logs (~1-10GB)
• cvat_inmem_db: Redis persistence (~100MB-1GB)
• cvat_cache_db: Kvrocks cache (10-50GB)
• cvat_events_db: ClickHouse analytics (~5-20GB)

​

Kubernetes Storage

• Backend PVC: 20Gi (default, configurable)
• Kvrocks PVC: 100Gi (default, configurable)
• PostgreSQL PVC: Based on Bitnami chart settings

​

Network Ports

​

Environment Variables

Key environment variables for configuration:

• CVAT_HOST: Hostname for accessing CVAT (default: localhost)
• CVAT_VERSION: Docker image version (default: dev)
• CVAT_POSTGRES_HOST: PostgreSQL hostname
• CVAT_REDIS_INMEM_HOST: Redis hostname
• CVAT_REDIS_ONDISK_HOST: Kvrocks hostname
• ALLOWED_HOSTS: Django allowed hosts
• CVAT_ANALYTICS: Enable analytics (0 or 1)
• CVAT_BASE_URL: Base URL for the application

​

Next Steps

Choose your deployment method:

• Docker Compose Installation →
• Kubernetes with Helm →
• AWS Deployment →