Roadmap

ParqDB develops one narrow path at a time. Current priorities follow the unified embedded and client/server API RFC.

Current Foundation

• SQLite catalog with persistent Parquet source definitions.
• Immutable Parquet index publication over file, S3, and HDFS warehouses.
• IVF source, LVQ4, and LVQ8 postings with shared centroids.
• Squared-L2 and cosine search over float32 and float64 source vectors.
• Embedded DataFusion planning, filtering, projection, exact fallback, and SQL composition.
• Bounded metadata, planning, centroid, and decompressed Parquet page caches.
• Process-scoped query admission and cancellable managed Arrow streams.
• Portable synchronous and asynchronous session facades.
• Process-scoped native index build coordination.
• Bounded incremental Arrow IPC stream encoding and decoding.
• HTTP transport and Python ASGI server with incremental Arrow IPC, source URI authorization, remote index lifecycle, cancellation, and restart coverage.
• Atomic refresh and reachability-based orphan removal.

Next

1. Run the complete portable conformance suite against embedded and HTTP transports as the public surface grows.
2. Harden authentication, deployment guidance, and non-Python protocol tests.

Later

• Batch vector queries with physical grouping by selected clusters.
• Additional index families and quantization schemes.
• Iceberg writing and a transactional shared catalog.
• Browser and non-Python clients over the stable HTTP protocol.
• Distributed SQL integration through a standalone reference compiler rather than a Python compute-engine plugin framework.

These items are directions, not release commitments. A feature becomes part of the supported surface only after its implementation, conformance tests, and operational limits are documented.