Streaming and pipelines converge in ClickHouse, dbt expands adapters

ClickHouse 26.8: querying as a pipeline, serving as an API

The most substantive work this week landed in ClickHouse 26.8, where two blog posts outlined a broader shift in how the database is positioned. Pipelined SQL introduces a syntax for composing multi-stage queries as a readable sequence of transformations — closer to a streaming model than the nested subquery style ClickHouse users are used to. That's a meaningful ergonomic change for building complex analytical flows, and it pairs naturally with the other headline: building a streaming HTTP API directly in ClickHouse. The latter uses named handlers, typed parameters, pagination, framing formats, and access control to let you serve query results as an API without a separate application layer. Read together, these two features point at ClickHouse increasingly acting as both the compute engine and the serving layer for real-time analytics.

The stable and LTS channels also saw steady maintenance releases, including v26.8.2.7-lts, v26.7.6.57-stable, and v26.6.4.55-stable, though no single change in those stands out as a story on its own.

Managed Postgres gains control, MySQL CDC goes GA

Two operational upgrades rounded out the ClickHouse orbit. Scheduled upgrades in ClickHouse Managed Postgres give Scale and Enterprise users defined maintenance windows, which is the kind of control that matters when you're running a production database you don't fully own. And the MySQL CDC connector for ClickPipes reached general availability, bringing faster parallel snapshots, safer defaults, and infrastructure-as-code support to MySQL and MariaDB replication into ClickHouse Cloud. CDC GA is a signal that the ingestion path is considered production-hardened.

dbt-core 2.0.0: one more adapter, one more rc

The dbt-core 2.0.0 train accelerated this week with three releases, culminating in v2.0.0-rc.1. The headline feature across dev.33, dev.34, and the rc is Phase 1 Exasol adapter support, covering connection/auth, seed types, and the full macro bundle for materializations and snapshots. Databricks also gets query tags with automatic metadata and profile-level config, plus the ability to persist legacy constraints. A security fix for tarball extraction — symlink and hard-link vulnerabilities — ran through all three releases. This is a case where the pattern of near-daily dev releases is worth folding into a single mention, since the story is the cumulative approach to the 2.0.0 release candidate, not any individual dev number. The rc feels like the last major feature gate before GA.

Tinybird JSON goes GA

Tinybird made its JSON data type generally available, removing the private beta flag so any workspace can create and query JSON columns. That directly addresses a practical friction point for teams holding semi-structured or dynamic event data. The release also notes that organization pages now load for members without workspace access, a smaller but welcome access-control improvement.

Benchmarking the agents

Finally, ClickHouse's Agentic Analytics Benchmark is worth a look even if you're not running an analytics agent today. The team benchmarked 28 models against 201 real production queries, measuring correctness, cost, and speed, and released the harness openly. With so many teams experimenting with natural-language analytics, having a public, reproducible benchmark is a useful reference point for evaluating what these agents actually get right.