<p><strong>TLDR:</strong> Xata has open-sourced its cloud-native Postgres platform with copy-on-write branching and scale-to-zero, built on CloudNativePG and OpenEBS. Confluent Platform 8.2 lands native queue semantics directly in Kafka, no extra systems required. Trendyol published a candid post-mortem on migrating from Neo4j to Apache AGE inside Postgres, trading graph performance for operational simplicity. Core Flow AI moved from Supabase to PlanetScale and cut latency to sub-10ms P99 while halving costs.</p>
<hr>
<h2>Xata open-sources its Postgres platform</h2>
<p>Xata released its cloud-native Postgres platform under Apache 2.0 after running it as a managed service. The platform runs on Kubernetes using CloudNativePG and OpenEBS, and the headline feature is copy-on-write branching that can clone a TB-scale database in seconds. It also includes scale-to-zero compute, PITR backups, and a serverless SQL driver over HTTP/WebSockets. Worth noting: it requires a Kubernetes cluster and is explicitly not recommended for single-instance setups, so this is squarely aimed at teams building internal Postgres-as-a-Service platforms or ephemeral dev/test environments.</p>
<h2>Confluent Platform 8.2 ships native queue semantics for Kafka</h2>
<p>Built on Apache Kafka 4.2, Confluent Platform 8.2 makes Queues for Kafka generally available. You get elastic consumer scaling, per-message acknowledgment, and task-level processing controls without bolting on a separate queue system. Flink SQL also hits GA in this release, with declarative stream management via DDLs and multi-Kubernetes cluster support. The deprecation of Confluent Health+ in favor of Unified Stream Manager is the other thing to track if you’re running fleet-wide monitoring.</p>
<h2>Trendyol migrates from Neo4j to Apache AGE</h2>
<p>The Trendyol engineering team published a detailed account of moving graph operations from Neo4j Enterprise to Apache AGE, the graph extension for PostgreSQL. The short version: variable-length path queries in AGE bypass indexes and were producing 3-5 second response times, so they replaced them with iterative fixed-depth traversals parallelized via Go goroutines. They also consolidated multiple relationship types into a single discriminated HAS edge to simplify writes. The trade-off is honest — they gave up Neo4j’s raw graph performance to get unified infrastructure, CDC via logical replication, and predictable licensing costs.</p>
<h2>Core Flow AI cuts latency and cost moving from Supabase to PlanetScale</h2>
<p>Core Flow AI, running a top-50 gen AI consumer app with 10M+ users, migrated off Supabase in one week after repeated outages and no query-level observability. Post-migration numbers: 1.5x better performance at half the cost, P99 latency under 10ms, and P99.9 read replica latency of 0.2ms. PlanetScale’s Insights tooling identified slow queries in minutes and helped reduce CPU usage by 30%. The speed of the migration and the observability gap are the real story here — if you can’t see what your queries are doing, you’re flying blind.</p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Neon + OpenAI Codex plugin:</strong> Neon launched an official Codex marketplace plugin that connects to Neon databases via MCP, letting developers provision databases, create branches, run migrations, and query schemas directly from the Codex chat interface.</li>
<li><strong>Databricks Document Intelligence:</strong> Databricks announced composable AI functions (ai_parse_document, ai_classify, ai_extract) for document parsing pipelines, claiming a 16% average agent performance gain and 5-7x lower cost versus VLM-based pipelines on OfficeQA benchmarks.</li>
<li><strong>Cloudflare + PlanetScale integration:</strong> Cloudflare is consolidating PlanetScale Postgres and MySQL billing into the Cloudflare dashboard, with automatic Hyperdrive integration for connection pooling and query caching; unified billing launches next month with Postgres starting at $5/month.</li>
<li><strong>CrateDB 6.3:</strong> Adds ALTER COLUMN SET/DROP DEFAULT, CREATE TABLE … LIKE, DROP SCHEMA CASCADE, and unifies object store parameters for Azure, S3, and GCS under COPY TO/FROM commands.</li>
<li><strong>Snowflake CORTEX_SEARCH_BATCH:</strong> A walkthrough shows the new batch function classifying 138 raw signals against a 60-node taxonomy in roughly 5 seconds, with LATERAL joins and METADATA$RANK for confidence thresholds.</li>
<li><strong>SSMS SQL Projects preview:</strong> SQL Projects are now in preview in SQL Server Management Studio, enabling database-as-code workflows with source control integration and CI/CD via sqlpackage CLI and GitHub Actions.</li>
<li><strong>memweave:</strong> An open-source Python library for agent memory using Markdown files and SQLite with hybrid BM25 + semantic search via sqlite-vec — no vector database, no server, no Docker required.</li>
<li><strong>dbt on Databricks:</strong> Databricks published a case for running dbt on its lakehouse, citing 2,900+ customers and pointing to Delta Lake/Iceberg materialization, Unity Catalog column-level lineage, and cost tracking tied to specific dbt runs.</li>
<li><strong><a href="http://PGConf.BE" target="_blank" rel="noopener nofollow">PGConf.BE</a> 2026:</strong> Scheduled for May 5th in Haasrode, Leuven; registrations are open and the preliminary schedule is published.</li>
</ul>


Query optimization, schema design, and data store coverage across PostgreSQL, MongoDB, Redis, and more.

Databases Digest

A roundup of major database and data infrastructure news: Xata open-sourced its Kubernetes-based Postgres platform with copy-on-write branching and scale-to-zero under Apache 2.0. Confluent Platform 8.2 brings native queue semantics to Kafka via Queues for Kafka GA, plus Flink SQL GA. Trendyol shared a post-mortem on migrating from Neo4j to Apache AGE in Postgres, trading graph performance for operational simplicity and unified infrastructure. Core Flow AI migrated from Supabase to PlanetScale, achieving sub-10ms P99 latency and halving costs. Also covered: Neon's Codex MCP plugin, Databricks Document Intelligence, Cloudflare+PlanetScale billing integration, CrateDB 6.3, and more.

Xata open-sources its Postgres platform, Confluent 8.2 ships queue semantics for Kafka