BetaWeekly digests are a beta — we're trying something new. Feedback welcome.
ClickHouse's platform expansion leads a week of agent-ready data tooling
November 9–15, 2026
ClickHouse unified its enterprise AI data stack, shipping managed observability, a security SIEM, and Postgres GA alongside an Xata acquisition, while Loom brought repository-grounded memory to coding agents and Dagster and dbt-core tightened their pipelines.
The week at a glance
Replay the weekBiggest releases
- 01Dagster 1.13.30 drops Python 3.10 and adds cloud deployment copy-materializations
Dagster - 02dbt Labs makes the case for warehouse-first AI context to cut agent token costs
dbt Labs - 03dbt-core 1.12.7 fixes source freshness fallback and --cache-selected-only scope
dbt Labs - 04ClickHouse Unifies Enterprise AI Data Stack with Launch of Expanded Platform
ClickHouse - 05Announcing ClickHouse Observability - fully managed observability on ClickHouse Cloud
ClickHouse
ClickHouse broadens from database to platform
The week's dominant story was ClickHouse's bid to become more than an analytics engine. It launched an expanded platform serving over 5,000 customers, with new capabilities spanning Postgres, observability, and security. The centerpieces arrived alongside it: ClickHouse Observability is a fully managed monitoring service where you just send telemetry and get Observability Apps built from a prompt, and ClickHouse Security enters private preview as an agentic SIEM built on RunReveal, with AI agents and 100+ integrations running against your own service. Underpinning the AI story, ClickHouse Managed Postgres went GA with NVMe-backed storage natively connected to ClickHouse, and the Xata acquisition will add instant branching, cloning, and scale-to-zero. Langfuse also contributed the AI Gateway, an open source entry point that traces every model call, caller, and cost.
Two posts make the platform more approachable. Loom gives coding agents shared memory and a code index grounded in the repository, so sessions build on prior work rather than starting cold. And a new self-service Previews page lets anyone in ClickHouse Cloud discover and enable preview features, join waitlists, and automate the same via API. For lakehouse users, ClickHouse now connects to Amazon S3 Tables to query Iceberg data in place and write results back for dashboards and agents. Behind all this, the engine kept its cadence with six stable and LTS builds across the maintained branches — routine churn worth noting only as steady maintenance.
Dagster hardens its Python baseline and cloud tooling
Dagster dropped Python 3.10, now requiring 3.11 or later — a baseline shift worth checking against your deployment images. For Dagster+ users, the cloud CLI gained a deployment copy-materializations command that moves the latest asset materializations between deployments while preserving timestamps, useful for promoting state across environments. The Dagster+ MCP server also picked up tools to read and toggle schedules and sensors and manage asset partitions, extending agent control over orchestration. On the safety side, alert emails and Teams alerts now escape customer-provided values to prevent HTML injection, and the asset catalog stopped polling health data for assets where it isn't shown, easing load on large workspaces.
dbt cleans up freshness and cache scoping
dbt-core's week was about correctness in pipelines. Freshness checks now fall back to per-source checks when batch metadata computation fails, instead of silently skipping every selected source under the skip_nodes_if_on_run_start_fails default — a fix that matters because a bad batch run no longer means invisible stale data. The relation cache in compile, show, and docs generate is now scoped to the selected nodes' schemas so --cache-selected-only is actually honored, and dbt deps uses a git credential helper rather than embedding credentials in the clone URL. On the editorial side, dbt Labs made the economic case for warehouse-first AI context, reporting that storing governed agent context in the warehouse instead of retrieving it from 20+ MCP servers cut token costs by more than half.
Releases covered11
- ClickHouse Unifies Enterprise AI Data Stack with Launch of Expanded Platform (opens in new tab)
- Announcing ClickHouse Observability - fully managed observability on ClickHouse Cloud (opens in new tab)
- Introducing ClickHouse Security in private preview, built on RunReveal (opens in new tab)
- ClickHouse Managed Postgres goes GA and Xata acquisition brings agentic capabilities to the service (opens in new tab)
- Langfuse launches AI Gateway (opens in new tab)
- Introducing Loom: agent memory that keeps up with your code (opens in new tab)
- SET allow_preview_features = 1: Self-service previews in ClickHouse Cloud (opens in new tab)
- Bring high-performance analytics to your lakehouse with ClickHouse and Amazon S3 Tables (opens in new tab)