Apache Ossie enters incubation, dbt reaches 1.12

A new home for semantic interoperability

The most consequential story of the week is the Open Semantic Interchange's acceptance into the Apache Incubator, where it will operate under the new name Apache Ossie (Incubating). The specification, community, and mission remain unchanged, but the governance shift to Apache's foundation signals a maturation of the project's long-term trajectory. For anyone building data products that depend on portable semantic layer definitions, this is a vote of confidence in the standard's stability and vendor-neutral future. All references to "OSI" in code and documentation are now historical, so teams should begin updating their tooling and naming conventions accordingly.

dbt-core 1.12.0 lands with breaking changes and DAG fixes

dbt Labs shipped dbt-core v1.12.0 on Thursday, a named release that removes the experimental dbt login command and the bundled dbt-state plugin. The breaking changes also finalize flag defaults that were introduced in earlier 1.x releases, now enabled by default. Teams still on older 1.9 or 1.10 configurations should review their dbt_project.yml to ensure their flags are explicitly set. The release also makes MAXIMUM_SEED_SIZE_MIB configurable, giving teams more control over seed file size limits.

A day earlier, the 1.12.0rc3 candidate fixed a subtle DAG issue: dependencies declared inside overloaded UDF bodies are now correctly carried onto the root function node, preventing dropped edges in the DAG. The release candidate also reduced parse-time memory overhead through virtualized macros and lazy no-copy manifest views, which should improve performance for projects with large numbers of models.

Tinybird Classic sunset and ClickHouse's Node.js RowBinary library

Tinybird announced that Tinybird Classic will be sunset for Free and Developer plan workspaces, with automatic migration to Tinybird Forward beginning September 15, 2026. Forward remains the primary product experience, offering schema changes, local development, branching, CI/CD, and AI-assisted features. Enterprise plan workspaces are unaffected, but Free and Developer users should plan their migration now.

Over at ClickHouse, the team released @clickhouse/rowbinary, a Node.js library for reading and writing the RowBinary binary format. The library is notable for its dual nature: it works as a standard importable parser, but also ships as an Agent Skill — a bundled SKILL.md file that allows a coding agent to read the specification and write a custom parser rather than calling the library directly. This is an interesting approach to making binary format handling more accessible in agentic coding workflows.

Data engineering narratives

Two blog posts from Dagster and ClickHouse offer practical case studies. Dagster's engineering team wrote about classifying a million Snowflake columns in nine days using a tiered AI classification system with human review, orchestrated through Dagster. The post is a concrete example of how tooling can accelerate data governance workflows at scale.

ClickHouse published a deep dive on replacing a traditional HDB (historical database) with ClickHouse for financial ticker data, covering the architecture decisions that made the switch viable for historical market data serving. A separate post explored configuring huge pages in ClickHouse Managed Postgres, explaining the performance implications of memory page size for Postgres's shared buffer cache.

Dagster also released version 1.13.14 (core) and 0.29.14 (libraries), fixing two errors in the serverless I/O manager related to multi-deployment organizations and storage keys with special characters.