Remac (@getremac) header
Remac (@getremac) profile picture

Connect AI agents and applications to current business data for request-time context, vector synchronization, and data delivery.

Joined March 2026
Photo by Remac (@getremac)Photo by Remac (@getremac)Photo by Remac (@getremac)Photo by Remac (@getremac)
Pinned Tweet
Remac is a high-performance, database-agnostic transaction log processor. It turns native database transaction logs into real-time streams, replicas, audit records, recovery data, and replayable events.
1
91
An account changes from 'starter' to 'pro', then back to 'starter'. A later table query correctly returns 'starter'. It cannot tell you that either committed transition happened. Current state answers what exists now. Committed history answers which accepted changes produced it, and in what order. Replication, audit, billing, and event processing can need different parts of that history.
1
1
273
One native database log can support six functions: CDC, logical replication, heterogeneous replication, audit, point-in-time recovery, and REDO and event replay. Each function needs a different result. They still share questions about selection, order, delivery, resume, state, progress, and recovery. Remac's product model governs those functions through the seven guarantees of the Remac Contract.
1
24
Frequent polling is still polling. It can work well when current row state is enough, deletes produce a signal, and the allowed delay fits the interval. Tighter freshness and more consumers add queries, cursors, retries, monitoring, and repair work.
16
Transaction log processing should become a primary data-infrastructure category. CDC introduced many teams to the value of committed change. The same foundation can do much more when correctness and recovery are designed across the full path.
1
1
123
Remac does not replace a data warehouse, database proxy, message broker, workflow engine, or general stream processor. It supplies the transaction log processing infrastructure those systems can receive committed data from.
18
A database-to-search path can start with one polling job. Then come cursor rules, delete handling, retry handling, failure isolation, reconciliation, alerts, and recovery ownership. The boxes can stay the same while the operating work keeps growing.
22
A transaction log processor must prove more than successful writes. It must prove what it selected, what order it preserved, what each sink resolved, where processing can resume, and what durable state remains for recovery.
12
Every committed transaction leaves behind a record of what changed and in what order. That record exists because the database needs it to preserve its own guarantees. But once you can process that history reliably, it becomes useful outside the database too. You can build downstream systems from the same sequence of changes the database itself considers committed. That’s the layer we’re building at Remac.
1
112
Remac exists to make every committed data change immediately useful to every system that must act on it. Transaction log processing is the infrastructure between the commit and those downstream systems.
13
A common contract can define completeness, ordering, delivery, and recovery across databases. The evidence for those guarantees remains source-specific.
1
11
CDC moves current changes. Replication converges state. Audit preserves selected history. Recovery restores an earlier target. Replay applies retained events again. Different jobs. One transaction log foundation.
1
17
Change Data Capture converts selected committed changes into row-level events and delivers them to configured downstream sinks in real time. It is one logical transaction log function.
10
External transaction log processing can require: - source configuration - privileges - retention settings - a logical decoder or physical streaming path - a source-specific position model Native history is the foundation. Dependable processing is the product.
11
The transaction log is the database's ordered native history. Its exact contents and access rules depend on the database, version, configuration, and processing mode. A processor must respect those boundaries before it normalizes anything.
1
16
One useful way to find distributed data problems is to start at the commit. Which systems must learn about it? What happens if the second update fails? Where can the missing work be recovered? The answers define the real data path.
1
17
A commit proves what the database accepted. It does not prove that search, analytics, another database, a message broker, or an audit store received the result. Those systems need a dependable path from committed data.
1
2
310
Remac retweeted
For years, I kept solving the same problem in different forms: the database commits, but the message publish or downstream write after it can still fail. Outboxes, queues, triggers, debezium, etc helped. The retries, cleanup, monitoring, reconciliation, infrastructure, and separate recovery tools remained. Eventually, I stopped treating them as separate problems. That led me to build @getremac , a high-performance, database-agnostic transaction log processor. Remac is built to move fast without weakening correctness, progress, or recovery. Those requirements became the Remac Contract. The vision is simple: make every committed data change immediately useful to every system that must act on it.
1
1
345
Decoded logical changes feed CDC, update replicas, create audit records, and support replay. Native physical log data supports recovery. These jobs have different rules. The database transaction log is their common foundation. Together, they form transaction log processing.
1
1
42