What are the three tiers of the DeltaForge architecture?

A control plane that stores catalog, RBAC, storage credentials, audit log, scheduled jobs and the worker registry; stateless compute workers that run the native SQL engine and accept queries over HTTP; and your storage (S3, ADLS Gen2, GCS, MinIO or local disk) that holds Delta Lake and Iceberg tables.

Where can DeltaForge be deployed?

On cloud VMs (AWS EC2, Azure VMs, GCP Compute Engine), on Kubernetes using the bundled Helm charts, or on-premises on bare metal, VMware, or Docker Compose.

Does DeltaForge depend on Spark or a JVM?

No. The control plane and the compute workers are native binaries. There is no Spark runtime, no JVM warmup, no cluster manager to operate.

Which client interfaces can connect to DeltaForge?

Desktop GUI, VS Code extension, CLI, MCP server for Claude and Cursor, and an ODBC driver consumed by Power BI, Tableau, Excel, .NET, Python, and R. All interfaces speak SQL to a worker over HTTP.

Is there a managed or hosted version?

No. DeltaForge is software the customer installs and operates on infrastructure they already control. There is no managed service in the data path.

DeltaForge Architecture: Control Plane, Workers, Your Storage

Three components, one platform

Each tier has a clear responsibility and runs on infrastructure you operate

Control Plane

Stores the catalog (schemas, tables, columns, views, pipelines), users, roles, storage credentials, audit log, cron-scheduled jobs, and the compute-worker registry. Metadata only: no data moves through it.

Compute Workers

Stateless native execution engines. Each query runs on one worker by default. Add more workers for more concurrency. Workers register with the control plane on startup and accept queries over HTTP.

Your Storage

Delta Lake and Iceberg tables stored as Parquet files on S3, ADLS Gen2, GCS, MinIO, or local disk. No proprietary format, no lock-in. Compatible engines read the same files.

Client Interfaces

Desktop GUI, VS Code extension, CLI, MCP server for Claude and Cursor, and an ODBC driver for Power BI, Tableau, Excel, .NET, Python, and R. All speak SQL to a worker over HTTP.

Where it runs

Customer-installed on infrastructure you already control

Cloud VMs

Deploy on Azure VMs, AWS EC2, or GCP Compute Engine alongside your existing cloud infrastructure. Storage talks directly to ADLS, S3, or GCS.

Kubernetes

Run control plane and workers as containers in your cluster. Helm charts ship with the product. Add more worker pods to increase query concurrency.

On-premises

Bare-metal servers, VMware, or Docker Compose on-prem. Pair with MinIO or a NAS mount for S3-compatible storage inside your data centre.

Security and access

Governance enforced at the query layer, not bolted on after

RBAC

Built-in and custom roles with GRANT and REVOKE at schema and table level. Workers check permissions before executing any read or write.

Audit log

Every query, credential access, and permission change is logged with user, timestamp, and outcome. Supports SOC 2 and HIPAA audit evidence workflows.

Credential vault

Storage credentials (SAS tokens, access keys, service account keys) are stored encrypted in the control plane and injected into workers at query time.

Three tiers, your infrastructure