Docen
Docen Parse 2.1 is here

High-precision document intelligence.

We train document intelligence models for teams that need accurate parsing, extraction, and review across messy real-world documents.

Try the playgroundRead the docsTalk to us
90+ languagesTablesMathHandwritingReading order
DOCEN PARSE · LIVE VIEW
Public records · 1891 grant deed
HDR
COUNTY OF PINE HOLLOW · REGISTER OF DEEDS
H1
Indenture of Grant
TEXT
IMG
TEXT
SIGNATURE
STRUCTURED OUTPUT
document_typegrant_deed
recorded1891-06-04
grantorH. A. Whitfield
parcelNW¼ Sec. 12
CONFIDENCE0.994
01 / 03

Built for teams working with hard documents

Financial services
Healthcare
Legal & compliance
Public records
Life sciences
Logistics
Insurance
Research labs
Government
Education
[02]PROCESSORS

Five processors, one document pipeline.

Each processor does one job well. Compose them into a pipeline that fits your documents and the schema your systems expect.

ALL PROCESSORS
P · 01Parse

Convert

Turn scans, PDFs, and photos into clean Markdown, HTML, and JSON with layout and reading order kept intact.

P · 02Recognize

Customize

Tune recognition and output schemas to your document types, then lock the behavior in across a whole corpus.

P · 03Extract

Extract

Pull structured fields against a schema and get a citation back to the exact span each value came from.

P · 04Segment

Segment

Detect regions, tables, and section hierarchy, and keep them aligned across long, multi-page documents.

P · 05Evaluate

Eval

Score parsing and extraction against your own labeled documents before anything reaches production.

P · 06Reference

More processors

Browse the full platform, client SDKs, and the API reference for every processor.

[03]BENCHMARKS·PUBLIC

Measured on the documents that break other tools.

We publish the numbers we hold ourselves to — recognition, tables, extraction, and latency — against sensible baselines.

HOW WE MEASURE
BM · CER↓ better
0.42%

Character error rate on scanned pages

Docen OCR0.42%
Baseline OCR3.10%
BM · TABLE↑ better
98.7%

Table-cell accuracy on complex layouts

Docen API98.7%
Open parser91.4%
BM · EXTRACT↑ better
94.3%

Field-level F1 on schema extraction

Docen Extract94.3%
Prompt-only82.0%
BM · LATENCY↓ better
~2.1s

Median processing time per page

Docen API~2.1s
Self-host baseline~6.4s
[04]PLATFORM·CORE

Compose, deploy, and monitor in one place.

The platform around the models: pipelines you can shape, deployment you control, and the metrics that keep quality honest.

EXPLORE PLATFORM
C · 01
ParseLayoutExtract

Compose

Chain parsing, layout, and extraction into one pipeline and shape the result to the schema your systems expect.

  • Markdown, HTML, JSON, JSONL
  • Per-step configuration
  • Deterministic output shape
C · 02
Docen APICloudVPCOn-prem

Deploy

Call one API, or run the same models in your own cloud, a private network, or fully offline. The outputs match everywhere.

  • Managed cloud endpoint
  • Client SDKs
  • Self-hosted images
C · 03

Monitor

Watch accuracy, latency, and cost per document type, and catch quality regressions before they reach production.

  • Per-type quality metrics
  • Confidence thresholds
  • Regression alerts
[05]VERTICALS·USE CASES

Built for the documents your team actually handles.

From invoices to filings to archives, Docen holds up on the formats that trip up generic tools.

ALL USE CASES
Finance

Invoices & accounts payable

Read totals, line items, and vendor details off messy invoices at scale.

View use case Finance

10-K financial filings

Extract figures and disclosures from long filings with citations to the page.

View use case Healthcare

Policy segmentation

Split long policy documents into clean, addressable sections.

View use case Research

Scientific PDFs

Handle complex tables, figures, and equations without losing structure.

View use case Multilingual

Japanese text in tables

Recognize dense multilingual text and keep table structure intact.

View use case Operations

Spreadsheets with gaps

Parse spreadsheets with empty cells and irregular headers reliably.

View use case
[06]MANAGED BATCH·NEW

Hand off the backlog.

Managed batch runs a whole queue of documents end to end and gives you structured output back — no pipeline to build or babysit.

Point Docen at a queue and walk away.

Managed batch takes a backlog of documents and runs it end to end. Throughput scales up when the queue is deep and settles back down when it drains, so you pay for the work rather than idle capacity.

  • Per-document status and structured output
  • Automatic retries and back-pressure
  • Millions of pages without standing infrastructure
Read about managed batch
01
Queue
02
Route
03
Process
04
Structure
05
Deliver
[07]OPEN SOURCE·COMMUNITY

Open workflows first.

Portable outputs, documented APIs, and SDKs that drop into the stack you already run.

Built on open formats.

Docen speaks the formats the rest of your stack already reads. Outputs are plain, portable, and easy to diff, so parsed documents flow into search, retrieval, and analytics without a lock-in layer in the middle.

MarkdownHTMLJSONJSONLHTML tablesBounding boxes
Explore the SDKs
docen.py
from docen import Docen

client = Docen(api_key=DOCEN_API_KEY)

doc = client.parse("archive-1891.pdf")

print(doc.markdown)      # clean, ordered text
print(doc.tables[0])     # structured table cells
print(doc.confidence)    # 0.994
[08]DEPLOYMENT·MODES

Run it wherever your data has to live.

The same models and the same structured output across managed cloud, your own VPC, and fully offline environments.

TALK TO US
D · 01

Managed Cloud

The fastest way to start. Send documents to the Docen API and get structured output back, with capacity that scales with your queue.

  • No infrastructure to run
  • Usage-based pricing
  • Regional processing options
D · 02

VPC Deployment

Run Docen inside your own cloud account. Documents stay within your network boundary while you use the same models and outputs.

  • Runs in your VPC
  • Your keys, your logs
  • Private networking
D · 03

On-prem · Air-gapped

Deploy to hardware you control, including fully offline environments. Built for teams with strict data-residency requirements.

  • Offline-capable images
  • Data never leaves the host
  • Hardware-flexible
[09]START

Process document queueswithout infrastructure.

Start on the API in minutes, or bring Docen into your own environment. Same models, same structured output, at whatever scale you run.

Get startedTry the playgroundView pricing