Parallelize the initial load.Remove the single-table bottleneck.

Supermetal copies tables concurrently and splits large tables into parallel chunks, keeping available CPU cores busy. Once the snapshot completes, CDC continues from the captured source position.

Source tables
One binary
Target chunks
Employees
4.0M rows
Orders
1.5M rows
Accounts
800K rows
Supermetal./supermetal
ParallelismAll cores active
Employees

3.8M of 4.0M

Orders

1.4M of 1.5M

Accounts

760K of 800K

From four services to one.

Common Kafka-based CDC stacks split capture, transport, and loading across separately operated JVM services. Supermetal runs the complete replication path in one Rust process.

A common Kafka-based change data capture stack sends data from a source database through Kafka Connect, Kafka, and another Kafka Connect deployment before reaching the target. Schema Registry sits beside Kafka as a lookaside service, not an inline hop. Supermetal sends data from the source through one Rust binary to the target.

A common Kafka-based CDC stack

Source database
Kafka ConnectDebezium source
Kafka
Schema Registry
Kafka ConnectSink
Target

With Supermetal

Source database
SupermetalSingle Rust binary · Apache ArrowDurable buffer on your object storage
Target
  • 4 middleware services
  • 4 JVM services
  • Kafka serialization boundary
  • Intermediate broker in the data path
  • 1 process
  • No JVM
  • No intermediate serialization
  • No intermediate broker

Manage and monitor from the same agent.

The agent includes its own web console and REST API, with observability built in. No separate control plane to deploy.

  • Web console

    Create, inspect, and manage connectors directly from the agent, with live status and replication metrics.

  • API, Terraform, and MCP

    Every console action is available through the REST API. Manage connectors as code with the Terraform provider, or inspect and operate them from MCP clients.

  • Built-in observability

    Export metrics, logs, and traces over OTLP, or expose metrics for Prometheus.

Deploy it inside your own network.

Run the agent as a binary or container in your VPC or data center. Data moves between your sources and targets without leaving your network boundary.

Inside your VPC or data center, the Supermetal agent sits between your sources and targets so data does not leave the network boundary.
Your VPC or data center
Sources
SupermetalSupermetal agent
Targets
  • Dedicated compute
  • No public IP required
  • Persistent state on storage you control
  • Isolate agents by workload, SLA, or region

Run the same agent locally.

Connect a real source and target locally and reach your first sync in minutes.

The Supermetal installer completes, then the local server starts and asks for an encryption passphrase used to protect connection credentials.