Capability · Data modes

Two data-exposure modes, one cascade.

Schema-only is a data-minimization mode — only headers and up to three short sample rows ever leave your system, so a DPA usually isn’t needed. Full-data unlocks row-level AI and media convert, gated on the PHI entitlement.

mode: schema-only | full-data

What you get

Built for regulated workloads

AdaptivMapr runs the same cascade whichever mode you’re in — what differs is how much of your data is allowed to leave your system. Schema-only is the default data-minimization mode: only your headers plus up to three short sample rows (each clamped to 80 characters) ever cross the boundary, so raw records never leave you and a DPA usually isn’t needed. Every map — schema-only or not — draws a small flat fee from your prepaid token wallet. Full-data unlocks row-level AI and media convert; it’s gated on the PHI entitlement, and only the metered layer-5 LLM call is routed to a PHI-eligible provider, under X-PHI and X-Region so it stays in-region and BAA-covered.

Schema-only

Barely anything leaves

Headers plus up to three sample rows, each clamped to 80 characters — enforced at a single chokepoint (clampForSchemaOnly). Raw records never leave you, so a DPA usually isn’t needed. It’s the default, and for most tabular imports it’s all you need.
Full-data

Row-level AI, PHI-gated

Unlocks row-level AI and media convert. Gated on the PHI entitlement — a full-data request without it returns 402 — so the exposure is always an explicit, entitled choice, never a silent default.
Residency

In-region, BAA-covered

When full-data does call out, only the layer-5 LLM step goes to a PHI-eligible provider, under X-PHI and X-Region — so the gateway forces a PHI-eligible, in-region model under a BAA. The rest of the cascade still runs in-process.
Shared

Same cascade either way

Both modes run the identical five-layer cascade in-process. Full-data doesn’t change the mapping logic — it only changes what layer 5 is permitted to see, and shares the very same input clamp.

Comparison

Schema-only vs full-data

Schema-onlyFull-data
PriceFlat per-map fee (prepaid wallet)Same wallet + AI tokens; PHI entitlement (402 without)
DPA requiredUsually not — data-minimizedYes — inherited BAA coverage
What leaves your systemHeaders + ≤3 sample rows (≤80 chars)Row values, to the layer-5 model
Row-level AINoYes
Media convert (OCR/STT/ETL)NoYes (general content only)
Where the LLM runsa PHI-eligible provider (only leftover headers)a PHI-eligible provider, X-PHI + X-Region, in-region
Cascade layers 1–4In-processIn-process

Both modes share the same input clamp (clampForSchemaOnly) and the same in-process cascade. Full-data only widens what the metered layer-5 call is permitted to see.

FAQ

Common questions

Only your column headers and up to three sample rows, each clamped to 80 characters. That’s enforced at a single chokepoint in the parser, so schema-only logic can’t accidentally ship more. No full rows leave you, so a DPA usually isn’t needed.
Because full-data lets row-level values reach the model, that exposure should be a deliberate, contracted choice. It’s tied to the PHI entitlement so the BAA and in-region routing are in place first — a request without the entitlement returns 402 rather than quietly sending data.
No. The same five-layer cascade runs in-process in both modes. Full-data only changes what the metered layer-5 call is allowed to see, and it uses the exact same input clamp — so the behaviour you tested on schema-only carries over.
Only the layer-5 LLM call leaves, and it goes to a PHI-eligible provider with X-PHI and X-Region set, which forces a PHI-eligible, in-region model under a BAA. Layers 1–4 never leave your process, in either mode.

Ready when you are

Put schema-only vs full-data in production — without shipping raw records.

Spin up a key in minutes. Top up a $10 prepaid wallet — every map is a few tokens.

$10 minimum to start · pay only for what you map · PHI under BAA coverage
Schema-only vs full-data — AdaptivMapr — AdaptivMapr