- Source connectors read a system’s native change stream and feed it into the engine (the capture pipeline).
- Sink connectors write the engine’s output documents to a target store.
- The real-time bus carries published events for the delivery pipeline.
Supported today
This matrix is the implemented surface for the current release. The architecture
is pluggable, but a connector is supported only when it appears here and in the
release test matrix.
Source connectors
PostgreSQL
Logical replication → embed related rows into each document, with
field selection and 1:1 / 1:many joins.
Neo4j
CDC → fold a node’s neighborhood into one document with multi-hop,
bounded fan-out.
MongoDB
Change streams and resumable bootstrap for collections in replica sets or
sharded clusters.
MySQL / MariaDB
Row-binlog CDC with snapshot bootstrap and memory or SQL projection modes.
Kafka / Redpanda
Consume Debezium change envelopes or raw JSON with sink-gated offset commits.
Realtime brokers
NATS
Use NATS Core for live-only delivery or JetStream for acknowledged,
cursor-based replay.
Redis Streams
Use Redis Streams for cursor-based replay when Redis is already part of the
operating environment.
Sink connectors
OpenSearch / Elasticsearch
Bulk upserts with deterministic document IDs. The same engine handles
both (shared
_bulk API).Meilisearch
Continuous search-index sync with joined documents, deterministic
primary keys, and delete propagation.
Redis
Ordered, idempotent key materialization with string or RedisJSON values,
optional TTLs, and replica acknowledgement.