The high-performance transaction log processor

Change happens.
We deliver.

Every committed change in your database, delivered in order and at-least-once, anywhere it needs to go: analytics, search, services, your data lake, etc. From a single binary, no JVM, no Kafka Connect, no cluster to babysit.

Read the docs
PostgreSQL today · 4 more databases coming · self-host anywhere
Works with your stack
PostgreSQLMySQLsoonMariaDBsoonMongoDBsoonSQL ServersoonOraclesoonPostgreSQLMySQLsoonMariaDBsoonMongoDBsoonSQL ServersoonOraclesoon
Apache KafkaRabbitMQAmazon S3MinIOCloudflare R2PostgreSQLFileWebhooksQdrantRedisGoogle Cloud StorageAzure BlobDynamoDBMySQLsoonElasticsearchsoonSnowflakesoonBigQuerysoonPineconesoonKinesissoonClickHousesoonGoogle Pub/SubsoonRedshiftsoonApache PulsarsoonApache KafkaRabbitMQAmazon S3MinIOCloudflare R2PostgreSQLFileWebhooksQdrantRedisGoogle Cloud StorageAzure BlobDynamoDBMySQLsoonElasticsearchsoonSnowflakesoonBigQuerysoonPineconesoonKinesissoonClickHousesoonGoogle Pub/SubsoonRedshiftsoonApache Pulsarsoon
The status quo

Moving data shouldn’t cost you a team.

Today the options are heavy, expensive, or fragile. Here’s what data teams are actually living with.

Dashboards a day behind

Nightly ETL means decisions get made on yesterday’s numbers — and fraud is caught 12–24 hours too late.

A cluster to babysit

Debezium means a JVM, heap tuning, and offsets that drift. Every deployment mode requires JVM operations. A pipeline that needs a pipeline team.

Bills that punish growth

SaaS tools charge per row synced. Your costs scale with your success — a misaligned incentive by design.

Replication that drifts

pg_logical is Postgres-to-Postgres only — no fan-out, no transforms. Roll your own with polling or webhooks and changes land out of order, or not at all.

Disasters with no undo

Nightly backups can’t rewind a bad migration. Without point-in-time recovery and REDO, you can’t replay to the exact moment before disaster — you lose everything since the last snapshot.

Search that lies

Re-index crons leave Elasticsearch hours stale — users search for items that exist but can’t be found.

The collapse

Six capabilities. Two modes.

The industry sells these as six separate products. They’re not. They all start by reading the transaction log five decode it into row-level events, one archives the raw bytes for full-cluster recovery.

CDC
Logical Replication
Heterogeneous Replication
Audit Logging
Point-in-Time Recovery
REDO / Replay
↓  all of it is just reading the log  ↓
MODE 01 · LOGICAL DECODING

Parse the log into ordered, row-level events. CDC, replication, heterogeneous sync, audit, and logical PITR are all just sinks on this one stream.

postgres · mysql · mongodb · sqlserver · oracle
MODE 02 · PHYSICAL ARCHIVING

Stream or archive the raw log bytes for full-cluster recovery. File-based for higher RPO, streaming for seconds. No broker in the path.

file-based · streaming
What teams remac

One pipeline, a dozen jobs it retires.

Real-time analytics

Kill the nightly ETL. Dashboards reflect reality with sub-second lag.

Search index sync

New rows searchable in milliseconds. No re-index cron, ever again.

Microservice sync

One source, many sinks, in order. Replace N polling integrations.

Audit & compliance

WAL-level capture. 100% complete, cannot be bypassed by application code.

Data lake ingestion

Continuous JSONL to S3. Decommission the Airflow DAGs.

Vector DB sync

Embeddings stay current with the source. No stale RAG context.

Logical & heterogeneous replication

Stream one engine into another — Postgres→MySQL, Oracle→Kafka — table by table, transformed in-flight.

Point-in-time recovery

Archive decoded events and replay any table to any moment — logical PITR without a backup window.

REDO & event replay

Re-run a stream from any offset to rebuild a sink that fell behind — no re-snapshot of the source.

1
single binary
low
memory footprint
ms
latency
seconds
recovery RPO
0
dependencies

“Every line of production code exists to move data from point A to point B — without losing it, duplicating it, or reordering it.”

Remac Engineering

“Trust comes from reliability. Reliability comes from testing. A rigorous test suite is not overhead — it’s the product.”

Remac Engineering

“We handle other people’s data. A pipeline that silently drops events is worse than having no pipeline at all — because the operator believes replication is happening when it isn’t.”

Remac Engineering

“Six capabilities. Two underlying modes. One binary. We didn’t build six products — we learned to read the transaction log correctly.”

Remac Engineering

“The question isn’t whether your pipeline works when everything is fine. It’s what happens when things go wrong — and ‘things go wrong’ has far more surface area.”

Remac Engineering

“Every line of production code exists to move data from point A to point B — without losing it, duplicating it, or reordering it.”

Remac Engineering

“Trust comes from reliability. Reliability comes from testing. A rigorous test suite is not overhead — it’s the product.”

Remac Engineering

“We handle other people’s data. A pipeline that silently drops events is worse than having no pipeline at all — because the operator believes replication is happening when it isn’t.”

Remac Engineering

“Six capabilities. Two underlying modes. One binary. We didn’t build six products — we learned to read the transaction log correctly.”

Remac Engineering

“The question isn’t whether your pipeline works when everything is fine. It’s what happens when things go wrong — and ‘things go wrong’ has far more surface area.”

Remac Engineering
Building with AI?

Keep your vector DB in sync in real time.

The AI Pack adds an embedding transformer and vector-DB sinks no cron, no drift, no stale RAG context.

See the AI Pack →

Deploy once. Forget the pipeline exists.

Your data arrives where it needs to be, in order, with delivery guarantees and your team never thinks about it again.

Read the docs