Push heavy comparison work closer to the data.
Agent-Based Connections place local agents near the source and target databases. That design reduces network pressure, keeps sorting and hashing close to each engine, and scales better when the reconciliation scope becomes large or geographically distributed. It is also the only execution model that can run the Chunk Hash algorithm.
How it works
In agent mode, Compare coordinates the run while local agents handle part of the read, sort, and hash workload close to each database. This architecture lowers central data movement and helps when source and target engines behave differently under direct ordering and encoding rules. It is also the mode required by Chunk Hash, where agents exchange chunk signatures first and return the first mismatching decrypted chunk only when Compare needs an exact diff.
| Aspect | Agent-Based Connections |
|---|---|
| Topology | Local agents run near the source and target, while Compare remains the control plane. |
| Operational advantage | Lower network transfer, better scalability, and more stable behavior across heterogeneous engines. |
| Main constraint | Extra rollout, lifecycle management, and local infrastructure requirements for each agent. |
| Exclusive algorithm | Chunk Hash is available only here. It compares hash windows first and expands only the first mismatching chunk into a detailed diff. |
| Typical fit | Large tables, remote environments, or cases where direct central reads are too expensive. |
Currently supported databases
Agent-Based Connections currently support only PostgreSQL and Oracle.
PostgreSQL
Currently supported for agent mode
Oracle
Currently supported for agent mode
Best-fit scenarios
The reconciliation scope is large or long-running
Agent mode is the safer default when central reads would create too much data movement or extend runtime unacceptably.
Source and target sit in different network zones
Local agents reduce dependency on long network paths between the Compare host and each database endpoint.
Sorting and encoding behavior differs between engines
Keeping heavy preparation work close to each engine reduces comparison noise caused by platform-specific ordering differences.
You want the first exact difference without moving every row
Chunk Hash is agent-only and keeps payload low by comparing hashes chunk-by-chunk, then requesting full values only for the first differing chunk.
Deployment checklist
Place each agent near its database
Deploy agents where database access is local, low-latency, and operationally stable.
Validate secure communication paths
Confirm that the Compare control plane and both agents can exchange control traffic reliably.
Check agent resources and version alignment
CPU, memory, and agent version consistency matter once hashing and sorting move off the central host.
Run endpoint and agent health tests
Validate both database access and agent readiness before defining mappings and execution rules.
Installation and security model
Agent-Based Connections are enabled by installing dedicated agent software. The agent package is available for macOS, Linux, Windows, and Docker deployments.
| Aspect | Agent operating model |
|---|---|
| Installation method | Install the agent software on macOS, Linux, Windows, or run it as a Docker deployment. |
| Logical binding | Each agent is logically associated with exactly one database endpoint. |
| Communication security | Traffic between the agent and Compare is authenticated with a Bearer Token. |
| Operational result | The execution model stays explicit: one agent, one database, one controlled Compare link. |
Stay with Direct Database Connections when the workload is modest, the network path is simple, and the team wants the lowest operational overhead.