LoginSign Up
← Back to all sparks
W

Windmill

INFRA · APIS
Velocity8.8

Open-source developer platform to build internal tools and workflows from scripts

www.windmill.dev ↗

Windmill gave away the warehouse connectors and now runs dbt natively — the trial is the strategy.

workflow orchestrationdbtopen coreai sessionsdeployment lineageself-hosted
◆Current state
Two lines are moving at once. The platform line made dbt projects a first-class runtime, unified the deployment target on workspace lineage, opened Compare & Deploy to arbitrary target workspaces, and moved BigQuery and Snowflake out from behind the Enterprise license. The AI line took sessions to beta on by default and has been adding controls around them since — file attachments, visible web-search sources, artifact version history, and now a read-only plan mode that refuses anything that writes or deploys until you approve a plan.
◆Where it's heading
Windmill is positioning as the place a data team's existing work already runs rather than a system to be ported to: an unmodified dbt project drops in, its models become addressable assets with ref() lineage, and the warehouse languages needed to reach them are no longer paywalled. In parallel, the AI session work is maturing from capability to governance — the recent additions are all about reviewability and constraint, not raw autonomy. The deployment changes point the same way, collapsing configurable targets into a single derived answer.
◆Prediction
Expect the plan-and-approve posture to spread beyond session chats into the durable AI surfaces, and more warehouse-adjacent runtimes to follow dbt into the first-class treatment; Oracle and MS SQL remain the obvious Enterprise holdouts to watch.

◆Recent moves

  1. 5d ago

    Plan mode for AI sessions

    A read-only autonomy posture for AI sessions: the assistant may investigate but is refused anything that writes, runs, or deploys until it produces a plan you approve. It fits the pattern of wrapping the beta AI sessions in review controls rather than extending what they can do unattended.

    View source ↗
  2. 8d ago

    Nested filter groups and dotted paths in trigger filters

    Kafka and WebSocket trigger filters gain nested groups with any_of/all_of/none_of logic up to three levels, plus dotted-path addressing into nested objects. Existing flat filters keep matching exactly as before, and unparseable filters are now rejected at save rather than failing silently.

    View source ↗
  3. 9d ago

    Version history for AI session artifacts

    AI session artifacts now keep labelled version history with a picker, a stale-view banner, and read access for the AI itself, so an earlier revision can be restored by asking rather than rewritten. Another reviewability control layered onto the sessions beta.

    View source ↗
  4. 18d ago

    Run dbt projects as a first-class Windmill runtime

    ⚡ SPARK

    An unmodified dbt project becomes a Windmill script: dropped into a __dbt/ folder and pushed, it runs dbt build on your own workers with per-model progress, retry, row previews, its own editor, and models exposed as addressable assets carrying ref() lineage. This is the anchor of the platform arc — the reason the warehouse languages were unlocked.

    View source ↗
  5. 20d ago

    One deployment target, derived from the workspace lineage

    The workspace-level deploy-to setting is removed in favor of deploying into the parent workspace, so the target has one definition instead of two that could disagree. Job tags become lineage-aware at the same time, with $workspace resolving to the nearest ancestor that actually has workers.

    View source ↗
  6. 20d ago

    Compare & Deploy into any workspace

    Compare & Deploy accepts any target workspace rather than only the parent, computing a full diff across both and deploying selected items one way. Requires admin on both sides, and the Deployment UI settings move into the Dev workspace.

    View source ↗