v2.0.0

BemiDB is now PgStack: Postgres read replicas for Search and Audit.

This release introduces a new architecture and a single-container local setup, keeping indexing and queries outside your application database.

What’s new

• Search with BM25 keyword ranking, semantic search, and hybrid retrieval.
• Audit history with before/after values, queryable through SQL.
• Continuous synchronization through Postgres logical replication.
• Configuration through pgstack.yaml, with Audit enabled by default.
• Optional OpenAI embeddings and SCRAM password authentication.
• Docker images for Linux AMD64 and ARM64, including Apple Silicon.

See the README for setup and Search and Audit examples.

v1.8.0

• Add a data syncer for Dialpad (via SMS webhook subscriptions)
• Implement in-memory vector similarity search on columns with embeddings
• Recursively process all expressions in COALESCE arguments #53
• Allow re-creating a DuckDB instance
• Handle 404 errors in a server from S3 during syncs

v1.7.0

• Add support for pg_cancel_backend(pid)
• Add support for flush and close PG protocol messages
• Improve handling DISTINCT ON ... statements
• Truncate decimals that can't fit into Parquet

v1.6.0

• Add support for json_array_elements()
• Add support for logical types (JSON)
• Add more standard Postgres types into pg_catalog.pg_type
• Allow extended protocol queries without a sync
• Store columns' info in the catalog
• Log the number of currently active connections
• Improve handling JOINs with multiple FROM clauses

v1.5.0

• Add support for json_column ? 'key' operator
• Add support for jsonb_array_length(jsonb)
• Add support for jsonb_pretty(jsonb)
• Add support for to_char(timestamp, 'YYYY-MM-DD')
• Fix Attio pagination

v1.4.0

• Add support for ::JSONB type casting
• Add support for WHERE 'value' = ANY(column)
• Allow per-column permissions in runtime
• Deprecate PG SOURCE_POSTGRES_INCLUDE_SCHEMAS environment variable

v1.3.0

• Add a data connector for Attio (CRM)

v1.2.0

• Add support for dbt with materialized views #25
• Add support for pg_rewrite table
• Allow renaming materialized views

v1.1.0

• Add support for materialized views
• Assign default column names for json_column->'value' or json_column->>'value' with AS json_column_value
• Fix ISO datetime parsing with microseconds without timezone

v1.0.1

• Compact Iceberg tables with multiple Parquet files per manifest file
• Improve Parquet file statistics collection
• Handle unexpected Amplitude 404 errors
• Convert timestamptz to timestamp in UTC data type for DuckDB (DuckDB C API limitation)