Skip to content
Vibedata

What's new · v0.1.69

See what your data engineering agent built

Seven read-only views show what a Domain defines, what an Intent changed, where an asset comes from, how the last run went and what state the data is in.

An open work log on a powder field. Three tabs along its top edge read Built, Changed and Ran; the Ran tab carries a navy check. The right page reads: what it built, what changed, whether it works.
The agent's work, open for inspection.

Inspecting the work used to mean opening the repo, finding the run, and guessing whether the model or the data was wrong. Seven read-only views now answer those questions in the conversation, beside the change they describe.

  1. Ask what exists

    A Domain repository panel on a powder field lists four kinds of asset: dbt models and their tests, dlt resources, semantic models, and orchestration items. A navy chip marks a since commit.
    What the Domain defines.

    One view lists the dbt models, dlt resources, semantic models and orchestration items in the Domain's repository, with a since commit to mark what changed. Ask for one of them and the answer explains it, with its latest result.

  2. Ask what changed

    A flow on a powder field from This Intent to Changed assets to Affected downstream. The Changed panel lists models, resources and items, and a caption below it reads Change Class. The Affected panel lists models and semantic models.
    What the Intent changed, and who reads it.

    One view names what this Intent changed, its Change Class and who is affected. Another shows where an asset comes from and what uses it, so a change to one model does not hide its downstream readers.

  3. Ask whether it works

    A run history panel on a powder field. The last build passed, the previous build failed with its reason recorded, and the health check compares the data only with the checks the repository declares. The tools list_runs, describe_run and get_health answer each row.
    How the runs went, and what the data shows.

    Two views show how recent runs went and why the last build failed. A third compares the data only with the expectations the repository declares, so it reports what is known.

Before Open the repository, read the diff, find the failing run in the logs and guess whether the model or the data is wrong.After Ask the conversation. The card names the changed models, the affected assets, the latest run and the declared checks.

The views report facts. You decide what they mean.

Get started