Release v0.3.2

• Promoted Anthropic support into the official agora-etl-plugins bundle through the anthropic extra
• Added the agora.ai.providers entry-point for AnthropicProvider under the agora_plugins.anthropic namespace
• Raised the core compatibility floor to agora-etl>=0.3.3
• Updated bundle docs and package quickstart guidance to position Anthropic as an official completion and structured-output integration
• Added regression coverage for the official Anthropic provider path, including structured-output and unsupported-embedding behavior

Release v0.3.1

• Raised the core compatibility floor to agora-etl>=0.2.2
• Updated package quickstart guidance to call out DeliveryConfig(batch_flush_interval_ms=...) for long-lived Redis, Kafka, and PostgreSQL source pipelines
• Refreshed release metadata so the plugin bundle tracks the new long-lived worker observability and timed batch flush behavior in the core runtime

v0.3.0

• Raised the core compatibility floor to agora-etl>=0.2.1
• Updated plugin docs to highlight DeliveryConfig(batch_size=100) for Redis, Kafka, and PostgreSQL source pipelines on the linear lane
• Documented that BatchMiddleware now works correctly with plugin sources that emit one record at a time, matching the agora-etl 0.2.1 runtime fix
• Added regression coverage for running a plugin source through BatchMapMiddleware on the linear lane

Release v0.2.2

• Reduced Kafka producer batch overhead by serializing synchronous batches eagerly and skipping the retry wrapper on the success path
• Restored Kafka send retries for custom retry policies that handle non-KafkaError transient failures
• Fixed async callable serializer detection so serializer objects with lifecycle hooks still serialize to bytes correctly
• Reduced Kafka source checkpoint churn by caching checkpoint payloads until processed offsets advance
• Improved Redis set batch throughput by switching TTL-free batch writes to MSET
• Added Redis stream consumer tuning knobs for batched acknowledgements and optional raw response handling
• Batched Redis stream acknowledgements and normalized byte-backed error payloads so shutdown, cancel, and fail-closed paths preserve cleaner replay metadata
• Reduced repeated xadd(maxlen=..., approximate=True) argument rebuilding in Redis stream sink batches
• Expanded Redis regression coverage for MSET batch writes and batched stream acknowledgements

Release v0.2.1

• Fixed PostgreSQL source install guidance to point to agora-etl-plugins[postgres]
• Hardened Redis DLQ iteration so bounded reads stop early and avoid loading the full DLQ index at once
• Improved Redis state TTL precision with millisecond-expiry writes and safer handling of already-expired keys
• Reduced Redis semantic dedup scan pressure by switching embedding lookups from full-set reads to incremental SSCAN batches
• Fixed PostgreSQL schema-adapter introspection for schema-qualified tables such as analytics.users
• Hardened Kafka source and sink startup/shutdown recovery so serializer or deserializer lifecycle is rolled back on startup failure and best-effort cleanup still runs during teardown errors
• Fixed Confluent schema registry requests to URL-encode subject path segments safely
• Expanded regression coverage for distributed coordination, Kafka recovery paths, Redis DLQ iteration, Redis TTL semantics, Redis embedding dedup scans, and PostgreSQL schema-adapter introspection