Schema API · Schema models

Put raw data to work

Schema is a Data Language Model. Input tables and databases as they are, across any number of sources, and get back the data understanding, how the sources relate, what is missing, and what can be predicted from it. No shared keys, no labels, no preprocessing.

Quickstart API reference

Key-lessCross-table mapping

AgnosticSector identification

Label-freeMissing-value imputation

InPOST /v2/run · quick run, multi-source, no shared keybash

curl -X POST https://api.schemalabs.ai/v2/run \
  -H "Authorization: Bearer $SCHEMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "tables": [
    { "id": "crm",   "columns": ["c0","c1","c2"],
      "rows": [["ana@x.io","555-0142","ACME"]] },
    { "id": "cards", "columns": ["a","b","c"],
      "rows": [["Ana K.","+1 555 0142","Visa"]] }
  ] }'

Outthe bundle, every output with its confidencejson

{
  "summary": { "tables": 2, "keys_used": "none",
               "records_linked": 1 },
  "tables": [ { "id": "crm",
    "sector": { "top1": {
      "name": "consumer financial services",
      "confidence": 0.89 } },
    "column_profile": [ { "name": "c1",
      "role": "phone", "pii": true,
      "role_confidence": 0.97 } ] } ],
  "cross_table_map": {
    "column_alignment": [ { "a": "crm.c1",
      "b": "cards.b", "attribute": "phone",
      "confidence": 0.96 } ],
    "entity_matches": [ { "a_row": 0, "b_row": 0,
      "confidence": 0.94 } ] },
  "imputation": { "filled": [] },
  "prediction": { "target": "cards.c",
    "task_type": "classification",
    "tier": "in_context" }
}

Start here

Pick your path

5 minutes

Quickstart

Get a key, run across sources, read the bundle. Then publish the same configuration as an endpoint.

Read

How it works

Introduction

What Schema returns, quick run versus endpoint, and the surfaces. The mental model in one page.

Read

REST

API reference

Every route with parameters, cURL, Python, and JavaScript samples, and full response examples.

Read

What you work with

The building blocks

A quick run for a one-shot answer, an endpoint that holds your data and serves a scored URL, data you connect once, and a held-out report on every endpoint. The platform and the API expose the same objects.

Quick run: the stateless call

POST /v2/run, or Quick run on the platform. One pass, full bundle back, nothing kept. For agents, pipelines, and one-shot analysis.

Read

minutes to live

Endpoints: state that follows your data

Pin data, get a held-out report and a stable URL. Refresh when the source changes, upgrade when a new base ships.

Read

Data: connections, datasets, snapshots

Upload, databases, cloud storage, APIs. Every sync mints an immutable snapshot; pinned data bills the cached rate.

Read

Held-out reports: the trust object

Every create, refresh, and upgrade emits a report with its split, seed, baseline, and weakest slice. Every number carries its provenance.