# Extract

> Extract is a low-latency API that parses documents (PDF, PPTX, DOCX, and images: PNG, JPEG, WebP, TIFF, HEIC/HEIF, BMP) into structured data — text, tables, and figures — in a single call, with per-span bounding boxes. It has the best text accuracy of seven providers (87.5% on a 400-page human-labeled gold benchmark across 7 document types) and the lowest median per-document latency of every hosted provider tested (2x faster than most). Pricing is simple credits at $0.003/credit — Parse is 1 credit/page ($0.003/page), Parse + Extract is 4 credits/page ($0.012/page), all-in — with 1,000 free credits on signup and no card required. Built from YouLearn's production document pipeline, which has processed 70,000,000+ pages. HIPAA + BAA available; never trains on customer data. Sync documents are never stored (processed in memory); async uploads and results are deleted automatically after 3 days.

Base URL: https://api.extract.page · Auth: `X-API-KEY` request header · Limits: 500 pages / 150 MB per sync request, async batches up to 1,000,000 pages.

## Start here

- [Full LLM context](https://hanji.dev/llms-full.txt): the entire API reference, quickstart (curl/Python/TypeScript), response shape, limits, error table, pricing, benchmarks, and FAQ as one markdown file. Read this first.
- [Documentation](https://docs.hanji.dev): guides and full API reference.
- [OpenAPI spec](https://docs.hanji.dev/openapi.json): machine-readable description of every endpoint.

## API endpoints

- [POST /v1/parse](https://docs.hanji.dev): parse a document from a hosted URL (JSON body).
- [POST /v1/parse/file](https://docs.hanji.dev): parse an uploaded document (multipart).
- Compatibility aliases: `POST /v1/extract` and `POST /v1/extract/file` still work and return the same responses; prefer `/v1/parse` for new integrations.
- [POST /v1/extract/schema](https://docs.hanji.dev): pull typed fields out of a document with a JSON schema.
- [Async batches](https://docs.hanji.dev/guides/batch): `POST /v1/files` + `POST /v1/batches` for jobs up to 1,000,000 pages.
- [Webhooks](https://docs.hanji.dev/guides/webhooks): opt in with `webhook: {"mode": "svix"}` on a batch to get a signed `batch.update` event when it finishes, instead of polling (Svix-wire-compatible signatures).

## Product pages

- [Home](https://hanji.dev): overview, benchmarks, and FAQ.
- [Accuracy benchmark](https://hanji.dev/bench): 400 human-labeled gold pages, 7 document types, 7 providers, scored character-level.
- [Compare head-to-head](https://hanji.dev/compare): run a document through Extract and a competitor live, side by side.
- [Demo](https://hanji.dev/playground): parse a document in the browser, no signup — output rendered with source bounding boxes.
- [Healthcare benchmark](https://hanji.dev/benchmarks/medical): results on medical documents.
- [Pricing](https://hanji.dev/pricing): free 1,000 credits, then $0.003/page for Parse (1 credit/page) and $0.012/page for Parse + Extract (4 credits/page); custom/enterprise tier with SLAs, dedicated regions, and HIPAA + BAA.

## Optional

- [Status](https://status.hanji.dev): live API status.
- [Contact](mailto:hello@hanji.dev): general and sales enquiries.