Möbius
Concepts

Custody modes

In place and Held — where your data lives, and what changes because of it.

Every table Möbius governs has a custody mode, decided by where the row data physically lives. Custody is independent of governance — certification, gates, PII masking, the audit trail, and every AI surface work identically in both modes. What changes is where the bytes are, and a small number of capabilities that follow directly from that.

In place — your warehouse holds the data

Data stays in your own warehouse (Snowflake, BigQuery, Postgres, and others). Möbius holds the governance layer around it: your semantic contract (column definitions), the data dictionary, schema profiling, certification history, and the audit trail — but never the row data itself.

  • A connector to your warehouse is required, because Möbius's certification predicates run as SQL against your data where it sits.
  • Once a table is certified, Möbius publishes a plain SQL view into your warehouse under a governed namespace — the certified view — so downstream consumers query a stable, governed object rather than your raw source table.
  • There is no write-back of individual rows: your warehouse is already the source of truth, so there's nothing to send back to it.

Held — Möbius holds the data

Data lives in Möbius's own store. Möbius holds everything an in-place table holds, plus the actual records.

  • No connector is required — certification predicates run against Möbius's own data directly.
  • Certification promotes records through an internal quarantine-to-live lifecycle as they pass each gate.
  • You can export certified data from Möbius back out to your own warehouse, under a namespace you configure — the reverse direction of an in-place table's arrangement.

Neither mode is "the" product

Some work is better suited to an in-place table — data never moves, and governance reaches directly into your existing warehouse. Some is better suited to a held table — holding the data enables row-level work like write-back, live previews, and drill-downs. Most workspaces use both: some tables are governed in place in a warehouse, others are brought into Möbius for deeper interactive work. Certification, the audit trail, PII masking, and the AI surfaces (schema understanding, dictionary suggestions, chat) apply the same way regardless of which mode a given table uses.

What's mode-limited, and why

A handful of capabilities are genuinely tied to where the bytes live, and Möbius surfaces the limit at the point where you'd hit it rather than after you've done the work:

  • Row-level authoring — decision views, dashboard drafts, and similar constructs that read or write individual records — apply to held tables. An in-place table's rows live outside Möbius, so there's nothing local to author against.
  • Import and data-profile refresh — uploading new rows, or refreshing sample-derived profile data — apply to held tables. An in-place table's profile is built from your warehouse directly.
  • The rejected-rows queue — only exists for held tables, because only a held table runs a local row gate with somewhere to hold what it rejects.

Everything else — the description, PII classification, masking strategy, ownership, and notes on a column in your data dictionary — is authored identically in both modes, and Möbius's AI features reason about a table's meaning from the same dictionary and schema profile either way.

What the product itself prints

The names above are how this page and the rest of the docs describe custody. The CLI, the audit trail, and API responses still emit the underlying identifiers — recognize them when you see them:

What you read hereWhat the product emits
In placeMode A, custodyMode: NO_CUSTODY
HeldMode B, custodyMode: CUSTODY

On this page