Skip to content

Core concepts

Five ideas cover almost everything you’ll do with the API.

A model is a document type plus the schema you want back (the same thing the app UI calls a parser). You parse against a model id. Prefer to skip schema design? Use schemaless extraction (POST /v1/extract) and LangParse infers the fields for you.

Parsing returns a documentId immediately (HTTP 202). Poll the document until status is processed — no long-held connections. The status moves through queued → processing → processed (or failed).

Each extracted field comes back with a confidence of high, medium, low, or flagged, plus bounding boxes — so you can auto-accept high-confidence values or route the rest to human review. This is how you measure accuracy rather than assume it.

Classify an incoming file and split multi-document PDFs, then send each part to the right model automatically — so a mixed upload becomes cleanly separated, correctly-typed results.

API keys carry scopes:

  • parse — submit documents and read results (new keys are parse-only).
  • manage — configure models, routers, and sources.
  • admin — organization administration.

Use a separate key per environment so usage is attributable and revocation is surgical.