Skip to content

Database Overview

Forge databases are services. They live inside projects, have deployments, variables, logs, data tools, settings, domains where applicable, and backups.

Forge can provision the following database and storage engines:

Engine Default port Category Notes
PostgreSQL 5432 Relational Supports logical replication and backups.
TimescaleDB 5432 Time-series PostgreSQL-family with TimescaleDB extension handling.
MySQL 3306 Relational Supports browsing, SQL console, row editing, backups, and restore.
Redis 6379 In-memory Supports key browsing/editing and backups.
MongoDB 27017 Document Supports collection browsing/editing and backups.
ClickHouse 8123 Analytical Supports browsing and SQL console.
Qdrant 6333 Vector Vector database for AI embeddings, semantic search, and RAG pipelines. Supports collection browsing and backups.
MinIO 9000 Object storage S3-compatible object storage. Use for file uploads, static assets, or as a drop-in replacement for AWS S3 in development. Bucket browsing, console UI on port 9001.

Qdrant is a high-performance vector search engine. Use it for:

  • Storing and querying AI embeddings (OpenAI, Cohere, etc.)
  • Semantic search and recommendation systems
  • RAG (Retrieval-Augmented Generation) pipelines
  • Similarity matching across text, images, or any vectorized data

Forge provisions Qdrant as a Docker container with a persistent volume. Variables QDRANT_URL and QDRANT_API_KEY are autogenerated and available to app services in the same project.

MinIO is an S3-compatible object store that runs on your own infrastructure. Use it for:

  • File uploads and user-generated content
  • Static asset hosting
  • Development S3 replacement (no AWS bill)
  • Backup target for applications

Forge provisions MinIO with a web console on port 9001. Variables S3_ENDPOINT_URL, S3_ACCESS_KEY_ID, S3_SECRET_ACCESS_KEY, and S3_BUCKET are autogenerated and available to app services in the same project. The console is accessible at the service’s public hostname on port 9001.

When you create a database service, Forge generates engine-specific variables such as POSTGRES_DB, POSTGRES_USER, POSTGRES_PASSWORD, REDIS_PASSWORD, MONGO_INITDB_ROOT_PASSWORD, QDRANT_API_KEY, MINIO_ROOT_USER, or MINIO_ROOT_PASSWORD.

App services in the same project can use database variable suggestions instead of manually copying generated values.

Database deployment ensures the Docker image, container, persistent volume, environment variables, runtime network, selected public-access route, and engine-specific settings are present.

Database services are not hot-swapped like app services. They use persistent volumes, and redeploying may briefly interrupt clients.

Provisioning, data tools, backup, and restore availability can differ by engine, server, and Forge Agent version. Forge checks the capability reported by the exact target before allowing an operation.

The availability shown on the service is authoritative. A blocked or unavailable action must not be inferred from another service or bypassed. In particular, connected-server recovery is available only when that Agent reports a qualified recovery capability for the engine.

When a root domain is configured, Forge can create public database hostnames after public access is enabled. The hostname and host port are shown in the database service settings.

Use public access for admin tools, migrations, or external clients that cannot join the Docker runtime network. Keep it disabled when everything talks privately from app services on the same server.

Database services can expose:

  • Table or collection browsing.
  • Row, document, or Redis key editing where supported.
  • SQL console for relational engines.
  • Qdrant collection browser and vector search console.
  • MinIO bucket browser via the web console (port 9001).
  • Data import for PostgreSQL-compatible and Redis services.
  • Backups and restores for supported backup engines.

The Backups tab shows whether the exact service and target support backup and restore. PostgreSQL, TimescaleDB, MySQL, MongoDB, Redis, Qdrant, and MinIO have engine-specific backup formats where the required target capability is available.

ClickHouse backup and restore are not currently available. A successful backup is not recovery proof until the recorded isolated restore test also passes.