This release closes most of the gaps between Forewarden and the dbt platform you may be comparing it with. Here is what is new, grouped by what you do with it, and where the limits are.
Environments and engines
Each environment now sets its own target name, so target.name in a macro returns what you expect, and extra profile keys such as a Snowflake query tag or a role. Environment variables work at project, environment, job and personal scope with dbt's precedence, and keys that start with DBT_ENV_SECRET_ are encrypted and scrubbed from logs. An environment can sit on a release track (v1 Latest or v2 Stable) or stay pinned to one engine, and a developer can override the engine for their own Studio work.
dbt v2, the open source dbt-oss runtime, is now a second engine beside dbt Core 1.12. A guided upgrade runs a readiness report, applies dbt-autofix on a branch, parses with both engines, switches one environment and compares the two runs node by node. Some warehouses are still experimental in v2: Postgres, SQL Server, Spark and Athena. Trino, Oracle, Teradata, Fabric and Synapse have no v2 adapter yet. Scheduled runs also keep dbt's partial parse cache between runs.
Runs that do less, and tell you more
A job can now build only what changed. The planner compares each model with its last successful build and skips the ones that would come out the same, and the run page says why for each node. On Snowflake, BigQuery and Databricks, cost insights read the warehouse's usage views and attribute credits, bytes billed or DBUs to runs, jobs and models. We verified cost collection against recorded data, not live accounts.
Notification channels can watch particular models and tell you when one fails, is skipped or goes stale against its freshness threshold. Status tiles show a model's health on a dashboard. Project docs are hosted behind your project roles, and a project can publish public models for another project to ref(), with access rules checked before the run.
A sharper editor, and AI that asks first
Studio now has hover, go to definition, find references, rename that rewrites every ref(), star expansion, a check for columns that do not exist upstream, dbt parse on save, YAML validation, CTE preview, sqlfmt and Prettier, an open pull request link, and Python models on warehouses whose adapter supports them. We have not yet run a Python model against a live Snowflake account.
The AI Data Engineer is still read-only by default. A project admin can turn on write mode, and then every file edit, dbt command, commit or pull request waits for your approval on a card that shows the exact change. A SKILL.md in your project is added to AI prompts. Amazon Bedrock and Google Gemini join the provider list. A staging model generator writes a source's staging SQL and YAML without any AI at all. AI stays off until an admin adds a provider key, and it is still in preview.
APIs, BI tools and Terraform
The REST API now covers administration, and a metadata API serves models, columns, tests, sources, lineage and metrics. There is an OpenAPI document, a small forewarden CLI, an Airflow example, and 18 MCP tools including run_dbt_command. A Terraform provider is built and tested, but it is not on a public registry yet.
Saved queries run by name, and exports write them as tables. An optional Postgres-compatible SQL endpoint lets BI tools read metrics with an ordinary driver. It is off by default and not exposed publicly, and we have tested it with psql and psycopg, not with Tableau or Power BI themselves. Boards can also be written as YAML in your dbt project and synced into the Lakehouse.
Identity, git and warehouses
Single sign-on now works with any OpenID Connect provider, including Okta, Keycloak and Google, set up in Admin. SAML is not built. Read-only seats are not billed. Studio commits can be signed with a per-user SSH key, and CI results post commit statuses to GitLab and Azure DevOps as well as GitHub; those two were checked against recorded requests, not hosted projects.
DuckDB and Apache Spark bring the count to 15 warehouses. DuckDB has run end to end, so six are now tested. Spark is in preview: profile and connection checks pass, but we have not run it against a live cluster. The help also has a setup guide for each warehouse, a Debugging runs page, and a Moving from dbt Cloud guide. Snowflake and BigQuery OAuth and private connectivity are not built.