Job & App
Lite first, full when you need it.
Both run in your workspace and use the same Unity Catalog tables. The App is the Job plus a workbench for people.
| AlloyMDM Job (lite) Built, not yet run on a workspace | AlloyMDM App (full) Designed, not built | |
|---|---|---|
| What it is | A task you add to any Databricks job | The Job plus a steward workbench |
| Setup | A YAML file and a job task | Setup wizard in the app |
| Review pairs | Write confirm / reject rows to the decisions table with SQL | Side-by-side evidence and one-click decisions |
| Change rules | Edit YAML or the rules table, run a preview | Rules editor with preview |
| Access control | Unity Catalog permissions on the tables | Unity Catalog permissions through user authorization; no roles defined in AlloyMDM |
| Ongoing cost | Compute for each run only | Run compute plus an always-on app and a SQL warehouse |
| Best for | Data teams that want golden records in a pipeline | Teams with stewards reviewing matches every day |
Fits where you already work
- Any job: run it after your gold tables refresh, in a Databricks Workflow or any scheduler that can run a Databricks task.
- Your own schema: output goes to a separate schema, so your gold tables are never modified.
- Several domains: accounts ship first; templates exist for customer, person, product, vendor and location.
Known limits today
- One engine on one machine. It refuses to run above a configurable record cap (500,000 by default) rather than run for hours, and it slows down on very crowded data.
- Full refresh on each run, not incremental.
- Decisions and rule changes take effect at the next run.
- Larger volumes will need a parallel engine, which is possible but not built.