doc-ai
Installation
SKILL.md
Document AI — Sarvam Vision
Live in-chat document extraction → sarvam-mcp (
sarvam_tools_vision_extract). This skill = SDK code.
Translating a document into another language? → translate/document — a different API (
client.document_translation.*), not this one.
[!IMPORTANT] Auth:
api-subscription-keyheader — NOTAuthorization: Bearer. Base URL:https://api.sarvam.ai(NOT/v1— that prefix is only for the OpenAI-compatible chat endpoint)
What this API does
Sarvam Vision is a 3B-parameter document-intelligence model, exposed through the Document AI API (SDK group: doc_ai / docAi). Two job types share one lifecycle (create → poll → fetch):
| Job | What it returns | Best for |
|---|---|---|
Digitise (doc_ai.digitise) |
Full-page OCR: text + layout + tables, as HTML, Markdown, or JSON | Archival, search, RAG ingestion |
Extract (doc_ai.extract) |
Only the fields you define, as structured JSON/CSV/XLSX | KYC, invoices, forms |
Supports PDF, PNG, JPG, and ZIP (flat archive of page images) across 23 languages (22 Indian + English).