AI agents, postgres migrations, and asset-aware orchestration

Agentic analytics and infra-from-CLI

ClickHouse had a defining week in AI and developer experience. The company announced the public beta of ClickHouse Agents, a fully managed agentic analytics service powered by Claude. After running it internally for more than a year, ClickHouse is now offering a no-code interface for building agents that can query, visualize, and alert on data directly inside ClickHouse Cloud. At the same time, ClickHouse became available in Stripe Projects, meaning you can now provision a ClickHouse or Postgres service on ClickHouse Cloud with a single command from the Stripe CLI — handing working credentials to your environment and handing off to the new clickhousectl CLI tool. For teams already using Stripe for payments, this dramatically lowers the friction of spinning up analytics infrastructure.

Postgres migrations and benchmarking

On the data movement front, ClickHouse introduced Postgres to Postgres ClickPipes, a fully managed migration service now in public beta. The tool addresses the pain of migrating Postgres between providers — something that usually involves manual dump-and-restore workflows or third-party replication tools — with minimal downtime and operational overhead. Separately, the team open-sourced ClickCannon, an internal benchmarking framework originally built for observability workloads. It replays real query traces, simulates concurrent users, and measures performance at scale, giving teams a more realistic way to size ClickHouse deployments than synthetic benchmarks.

Dagster's asset hierarchy and selection grammar

Dagster's 1.13.9 release (core) and 0.29.9 (libraries) introduced several quality-of-life improvements for large-scale orchestration. Asset group names can now contain / separators to define hierarchical groups (e.g. marketing/ads), which render as nested groups in the asset graph and support wildcard selections like group:"marketing/*". The release also adds an is: filter to the asset selection syntax, letting you select assets by type — is:external, is:materializable, and so on — making it easier to target subsets of a large asset catalog without memorizing naming conventions. Meanwhile, a new blog post on making the architectural case for Dagster maps out an Orchestration Maturity Model, contrasting the implicit dependency chains of job-centric systems with Dagster's asset-aware approach to freshness, lineage, and quality at enterprise scale.

Tinybird's secret snapshotting

Tinybird shipped a behavioral change that matters for deployment workflows: Pipes that use secrets now snapshot their values at deploy time and continue using those snapshotted values even if the underlying secrets change. Previously, a secret change would take effect immediately on every pipe invocation, breaking the mental model of Forward deployments. Now, new secret values require a redeployment to take effect, aligning the behavior with how teams expect immutable deployments to work.

ClickHouse stable releases

ClickHouse shipped three maintenance releases: v26.4.4.38-stable, v26.3.13.31-lts, and v26.5.2.39-stable. These are routine patches for bug fixes and stability improvements across the stable and LTS tracks.