Installation
ContextCrate can run as a compact standalone application or as independently scalable distributed roles. Choose the smallest deployment that fits the workload, but use Kubernetes for production.
| Method | Use it for | Production suitability |
|---|---|---|
| Docker | Fast local evaluation from a published image | Small standalone deployments |
| Docker Compose | Local development, standalone evaluation, or distributed evaluation | Standalone only; distributed Compose is evaluation-only |
| JAR | Local Java development and direct JVM operation | Small standalone deployments |
| Kubernetes | Managed, resilient, or scalable installations | Recommended |
All methods require a non-default CONTEXTCRATE_ADMIN_PASSWORD. The default initial administrator is admin@contextcrate.local; change the email with CONTEXTCRATE_ADMIN_EMAIL when needed.
Deployment profiles
The standalone profile runs every pipeline stage in one process using H2, filesystem artifacts, and Lucene. Persist /app/data because it contains application data and the local embedding-model cache.
The distributed profile separates the control plane, source workers, browser crawler, parser, and indexer. It uses PostgreSQL, RabbitMQ, S3-compatible object storage, and OpenSearch. Use independently operated backing services, durable volumes, Kubernetes Secrets, and an Ingress with TLS for production.
See Deployment Notes, Configuration, and Operations for topology, environment variables, backups, and scaling guidance.