01Parse
Clean Markdown from scans, photos and PDFs, in reading order.
1st on olmOCR-Bench Ahead of
Claude Opus 5 and 
Mistral OCR4 02Extract
Only the fields you ask for, each with its page. Never guessed.
5th of 51 on ExtractBench Ahead of Claude Opus 5.5 and 
GPT-6 Astra 03Classify
Your own document types, each with a confidence score.
04Split
Finds where each document starts and ends in one long PDF.
05Workflows
Chain every step on one canvas and run it as one API.
01 · Extract
Extract reads every page of a long file, so values deep inside it come back as reliably as those on page 1.
See Extract02 · Parse
Parse reads print and handwriting in English and 22 Indian languages, including faded scans and stamped pages.
See Parse03 · Parse
Parse keeps rows, columns and headers in place on dense, ruled tables like registers, ledgers and statements, and returns them as Markdown or HTML tables.
See Parse04 · Extract
If a value is not in the document, Extract returns it empty. Every value it does return comes with its page and a confidence score.
See ExtractEvery operation is a job. Submit the file, poll its status, then fetch the result.
curl -X POST https://api.sarvam.ai/doc-ai/v1/job/parse \-H "api-subscription-key: $SARVAM_API_KEY" \-F file=@./sample-document.pdf \-F language="en-IN" \-F output_format="md"# {"job_id": "<JOB_ID>", "status": "pending", ...}
Returns a job_id. Poll /job/{job_id}/status, then download the Markdown from /download-url.
Read the guidePrices are per page, before tax. Volume discounts apply as usage grows.
Parse
₹0.50 per page
Extract
₹1 per page
Classify
₹0.50 per page
Split
₹0.75 per page
Start from the files your team already gets, and end with clean data your systems can use.
Statements, applications and property papers arrive in one bundle. Every value the underwriter needs comes back with its page.
Policy schedules, claim forms and hospital records are read before anyone opens them. Anything uncertain goes to a person.
Bills, invoices and purchase orders all come out in the same format, so reconciliation starts from clean data instead of paper.
Encumbrance certificates, registers and land ledgers come back as clean data, with values kept in their original script.
Agreements on stamp paper, even skewed or photocopied ones, come back with their parties, dates and stamp details.
Forms filled in by hand, in Hindi or English, come back as clean fields. A blank field stays blank instead of being guessed.
Not every team needs fields. Policies, circulars and case files are parsed whole, with headings and tables intact, so your search and AI assistants work from the full document.
Sarvam Cloud
Fully managed in India, and the fastest way to start
On-premises
Runs in your own data centre, so documents never leave it
Private cloud (VPC)
Runs inside your cloud account, under your network controls
Security & privacy
Process loan files, claims and contracts on ISO 27001:2022 certified, SOC 2 Type II attested infrastructure.
Your documents remain yours. We do not train on them, and you decide how long they are retained.
Bring your most complex document. See it structured in minutes.
Bring your most complex document.
See it structured in minutes.
