Documentation

Documents — parse (beta)

POST /v1/documents/parse — upload a PDF, scan or image and get cleanMarkdown back, transcribed by a routed vision model (defaultauto:vision — pin any vision-capable model with model). Built for the documents Indian enterprises actually hold: invoices, KYC, contracts, HR paperwork.

DPDP posture: processing is fully in-memory, zero-retention — document bytes and page images are never written to disk or logged; the response is the only output. Combine with data_policy: india_only and the parse never leaves India.

curl https://api.bharatrouter.com/v1/documents/parse \
  -H "Authorization: Bearer br-..." \
  -F file=@invoice.pdf

JSON alternative: {"file": "<base64>", "model": "auto:vision", "pages": 10}. Response: {"object":"document.parse","pages":N,"markdown":"…","model":…,"provider":…}.

Field	What it does
`file`	multipart file field, or base64 string in JSON. PDF, PNG, JPG. Up to 32 MB.
`model`	Vision-capable catalog id or an `auto:*` preset. Default `auto:vision`.
`pages`	Cap pages parsed (server max 50 in beta).
`data_policy` / `provider` / `exclude`	Same routing controls as chat.
`preset`	Bundled extraction schema: `invoice` (GSTIN, line items, CGST/SGST/IGST, totals) · `kyc` · `resume` · `contract`. Adds a validated `fields` object to the response.
`schema`	Your own JSON Schema for extraction (an object in the JSON body; a JSON string in multipart). One schema-feedback retry on invalid output, then `fields: null` plus a warning.
`output`	`markdown` (default) · `json` (fields only) · `both`.
`redact`	`true` → mask PII and secrets (Aadhaar, PAN, phone, email, cards, API keys…) in the returned markdown and fields — per request, independent of your org's guardrail settings.

curl https://api.bharatrouter.com/v1/documents/parse \
  -H "Authorization: Bearer br-..." \
  -F file=@invoice.pdf -F preset=invoice -F redact=true -F output=both

Metering: each page — and the extraction pass, when a schema/preset is given — is metered as a normal vision-model call (your BYOK key when routed there; platform routes at catalog rates). Roadmap: bulk parsing via/v1/batches, a console upload page, and an MCPparse_document tool.