We’re a lean team that likes to ship at incredibly high velocity. Here are the main features we’re currently working on.

​
In Progress

​
Analytics

  • A custom scan node for aggregates. This will allow “plain SQL” aggregates to go through the same fast execution path as our aggregate UDFs, further accelerating aggregates like COUNT, and SQL clauses like GROUP BY.

​
Write Throughput

  • Background merging. Improves write performance by merging index segments asynchronously without blocking inserts.
  • Data model optimizations. Reduces the storage overhead of index segments.
  • Pending list. Buffers recent write before flushing them to the LSM tree.

​
JOIN Improvements

  • Scoring and highlighting across JOINs. BM25 score and snippet functions can be used in JOIN queries.
  • Smarter JOIN planning for search indexes. Apply index-aware optimizations and cost estimation strategies when multiple BM25-indexed tables are joined.
  • Faster JOIN performance through predicate pushdown. Search predicates are selectively pushed down to relevant tables based on indexability and selectivity, improving JOIN query speed.

​
Improved UX

  • More intuitive index configuration. Overhaul the complicated JSON WITH index options.
  • More ORM friendly. Overhaul the query builder functions to use actual column references instead of string literals.
  • New operators. In addition to the existing @@@ operator, introduce new operators for different query types (e.g. phrase, term, conjunction/disjunction).

​
Long Term

​
Managed Cloud

  • Today, you can deploy ParadeDB either self-hosted or with ParadeDB BYOC. We are working on a fully managed cloud offering, with a focus on scalability and supporting distributed workloads.

​
Deeper Analytics Improvements

  • Push Postgres visibility rules into the index. This is currently a filter applied post index scan that adds overhead to large scans.
  • Evaluate more industry-standard OLAP tools. A new file format? Query execution library?

​
Vector Search Improvements

  • Postgres (and by extension, ParadeDB) uses pgvector for vector search. Contingent on demand and internal resources, we may investigate what improvements can be made to the known limitations of pgvector.

​
We’re Hiring

We’re tackling some of the hardest and (in our opinion) most impactful problems in Postgres. If you want to be a part of it, please check out our open roles!