Document parsing API

Documents in.Structure out.

Turn PDFs, scans, slides, spreadsheets, and images into Markdown and DocIR JSON for search, RAG, automation, and agents.

20 documents free · deterministic parsing · 200 MB · no card required

FY25 report12

Revenue by segment

Platform revenue rose to $82 million, mostly from seats added in the second half.

Services stayed near $41 million. Renewals covered the work that did not grow.

SegmentFY25Change
Platform$82M+31%
Services$41M+12%
01 / The parsing gap

Tables, column order, and page numbers drop out before the prompt.

A flat text extract drops tables, reading order, page boundaries, and the warnings you need when a result is wrong. DocParse keeps that structure on the parsed record.

Raw extractionContext lost

A wall of text your pipeline has to clean up later.

  • ×Multi-column paragraphs arrive in the wrong reading order.
  • ×Tables collapse into strings and headings lose hierarchy.
  • ×No page, box, parser, or confidence metadata for debugging.
DocParse outputContext retained

A normalized document layer ready for downstream work.

  • ✓Readable Markdown for chunking, search, and model context.
  • ✓DocIR JSON with pages, typed blocks, bounding boxes, and order.
  • ✓Manifests, warnings, lifecycle events, and stored artifacts.
02 / Built for the ingest layer

One parsing layer covers four workflows.

The files are already on the customer side. The product needs headings, tables, page numbers, and a route you can inspect.

01 — RAG

Knowledge & retrieval

Preserve headings, tables, page references, and reading order before chunking and embedding.

Chunk by heading and page
02 — PRODUCT

AI product ingestion

Accept PDF, Office, HTML, XML, CSV, and images through one API.

Skip a separate parser per format
03 — OPS

Document workflows

Normalize uploaded files before classification, review, extraction, routing, or human approval.

One record per uploaded file
04 — DATA

Search & archives

Turn mixed-format repositories into consistent, traceable artifacts for indexing and migration.

Index pages, blocks, and source ids

Route from the file

A digital PDF, a scan, and a spreadsheet take different routes.

Digital PDFs take a deterministic text-layer path. Scans and complex layouts can use OCR or layout vision. Office files keep their native structure. You get one API and a traceable route decision.

The free Trial validates deterministic parsing first. Image, Fast AI, and Layout-aware processing are enabled only for approved paid workspaces.

Digital PDF
text layer
Scans
OCR
Complex layout
vision
One normalized
output
01 / CONTROL

Inline for a proof. Jobs when the file is slow.

Use inline parsing for normal files or jobs, queues, retries, cancellation, and webhooks for slower work.

02 / TRACE

The route, warnings, and artifacts stay on the job.

Inspect routes, job events, warnings, parser versions, manifests, and tenant-scoped artifacts.

03 / BOUNDARY

Cloudflare-native by default.

External OCR is explicit and tenant-controlled. URL imports and webhooks can be restricted by host.

03 / Cloudflare DocParse API

Ship document features without operating a parser stack.

Send a file to one authenticated endpoint and receive clean Markdown plus layout-aware DocIR JSON. Start inline, then move the same integration to async jobs and webhooks as volume grows.

01

Create a scoped key

Open a free workspace, name the key for your backend, and copy the secret once.

02

POST the document

Upload multipart or raw binary, or import a permitted HTTPS URL.

03

Use the result

Read result.markdown for RAG and result.document when layout, blocks, and provenance matter.

First request · cURL
# Keep the key in your backend environment
export DOCPARSE_API_KEY="your_key"

# Parse now and save the complete response
curl -X POST "https://docparse.genedai.me/v1/parse?wait=true" \
  -H "Authorization: Bearer $DOCPARSE_API_KEY" \
  -F "file=@document.pdf" \
  -F "mode=both" \
  -o result.json

# Send Markdown to retrieval; keep DocIR for structure
jq -r '.result.markdown' result.json > document.md
jq '.result.document' result.json > document.docir.json
Prototype

One request, complete result.

Add ?wait=true for normal files and receive the job, Markdown, DocIR, and manifest together.

Production

Queue work without changing the input.

Remove ?wait=true, then poll the result link or receive a terminal webhook.

Control

Operate within clear boundaries.

Use scoped keys, quotas, idempotency, cancellation, artifact purge, and tenant-specific URL policies.

04 / Production controls

Keys, quotas, idempotency, and purge are part of the API.

A workspace can revoke a key, cap monthly jobs and bytes, and delete retained files without guessing what the parser stored.

#

Tenant-scoped access

Hashed API keys, explicit scopes, revocation, monthly quotas, and concurrent job admission.

↺

Operational safety

Idempotency keys, retries, dead-letter handling, cancellation, purge, and signed webhooks.

✓

Clear provenance

Original clean-room parsing code, permissive dependencies, and Apache-2.0 project licensing.

05 / Learn before you integrate

Product pages, workflows, comparisons, and the API quickstart.

Product

Understand the parsing layer.

Outputs, routes, lifecycle, access, cost, and retention in one product system.

Explore product →
Build for

Start from your workflow.

Architectures for RAG, AI agents, knowledge bases, search, and automation.

Choose a use case →

Check the parse on your own files.

The free workspace covers 20 text-based documents and 200 MB per month. Trial parsing is deterministic. Read the Markdown and DocIR before you rebuild the pipeline around it.

Create free workspace

20 documents · 200 MB
Deterministic parsing · no payment card