The document layer for evidence-ready enterprise AI.
Turn documents into structured, source-linked data for AI. Built-in, open-source Ethos checks citations against the original document.
DOCUSHELLFinancial report
q3-operations-report.pdf04 / 18
Q3 Operating Review
Revenue performance
Recurring revenue expanded across the enterprise portfolio, lifting quarterly performance above plan.
Revenue$12.7MYoY growth+18%Margin24%
Quarterly revenueQ1–Q3 · +18% YoY
Enterprise$8.4M+18%
Mid-market$3.1M+11%
Services$1.2M+6%
Structured summary · values remain linked to their source regions
PARSING THE DOCUMENT…
DETECTED · REVENUE TRENDEvidence linked04 / 18 · exact region retained
Layout-aware parsing
Deterministic output
Source-linked evidence
Temporary processing
PAGE CONTENT RECOVERED
The page becomes machine-readable.
OCR recovers visible content from the source document while its page boundary stays intact. Structure and evidence still need reconstruction.
- Visible text recovered
- Page boundary retained
- Structure still unresolved
Read page content→Structure document data→Verify source evidence
LAYOUT-AWARE DOCUMENT EXTRACTION
Plain text loses context.
Tables, labels, sections, and source locations can lose their meaning during basic extraction. DocuShell keeps this structure and connects every result to the original document.
Complex documents lose proof when flattened
Contracts, policies, manuals, filings, and reports collapse into loose text. Clauses, tables, definitions, and exceptions become easier to misread.
RAG needs source-backed citations
Return page context and coordinates so answers can point back to the source material—not retrieval text alone.
Workflows need stable evidence
Use deterministic JSON and Markdown for fields your application must route, review, cite, or audit.
SOURCE MAPS FOR CITATIONS
Parse once. Return source-backed structure. Let every AI workflow cite the page and region it used.
CORE CAPABILITIES
One parse layer. Every fact stays connected.
DocuShell turns document structure into application-ready output while keeping the path back to the original page.
01
Parse with structure
Preserve headings, paragraphs, tables, fields, figures, and reading order instead of returning a flat text dump.
02
Carry source context
Link extracted values to page references and regions so a reviewer or agent can inspect the exact supporting evidence.
1{2 "total_due": 8420,3 "currency": "USD",4 "source": { "page": 1, "bbox": [404, 178, 559, 244] }5}
03
Return usable artifacts
Return deterministic JSON, Markdown, HTML, and text—with Ethos evidence that connects important claims to the original document.
SOURCE-LINKED EVIDENCE
Find the exact source.Understand the context.Move forward with confidence.
A parsed answer is useful. A parsed answer that opens the supporting page, clause, table, or field is operationally trustworthy.
VERIFIED ANSWER
What is the liability cap?
Fees paid during the preceding 12 months.
Sourcemaster-services-agreement.pdf · page 7 · clause 8.2
Contract source
The surrounding document contains the relevant provision and its qualifying language. Review the highlighted region before moving the claim into an automated workflow.
Except for excluded claims, each party’s aggregate liability will not exceed the fees paid in the twelve months preceding the event giving rise to the claim.
page 7clause 8.2bbox [64, 192, 538, 361]
HOW IT WORKS
From raw document to reviewable data.
A controlled three-step path carries structure and source context from upload to the systems that need it.
01Connect a document
02Parse structure and context
03Ship source-linked output
STEP 01
Connect a document
Send a document to the Parse API or begin with the playground. Born-digital and scanned inputs enter the same controlled workflow.
DOCUMENT INPUT
Meridian Q4Performance review12 pages · 2.4 MBPOST
/v1/parsefiledocument
outputjson, markdown
Document attachedInput accepted Ready to parse
STEP 02
Parse structure and context
DocuShell preserves reading order, headings, tables, fields, page references, coordinates, and extraction warnings.
MERIDIAN / Q4PAGE 01
Q4 Performance Review
Revenue grew across all segments in Q4.
HEADINGCHARTTEXTTABLEHEADINGQ4 Performance Review
CHARTRevenue trend
TEXTGrowth summary
TABLEPerformance table
page 1 · 1 chart · 1 table · 184 text tokens
source map available
STEP 03
Ship source-linked output
Return JSON and Markdown with evidence anchors your application can inspect, cite, route, or withhold for review.
PARSE RESULT
revenue_q4"$1.67M"page1bbox[72, 168, 516, 298]SOURCE MAP ATTACHED
ETHOScitation grounded
AI agentAnswer with citation
RAG indexChunk + source map
Review queueOpen page 1
SECURITY
Security built for enterprise documents.
DocuShell validates inputs, isolates processing, delivers outputs through controlled workflows, and removes temporary files after delivery or within the one-hour cleanup window.
Hardened validationSchema checks, file signatures, size limits, and controlled parsing paths.
Ephemeral storageUploaded and generated documents are temporary processing artifacts.
Explicit outcomesPartial and review-required states remain visible instead of being converted into confident guesses.
ephemeral processing boundary
Validate
Process
Deliver
Purge
BUILT FOR DEVELOPERS
One request. Structured output with a path back to the page.
Start in the playground, then move the same Parse API workflow into an authenticated product or pipeline.
PLAYGROUNDmeridian-q4-performance-reviewRun parse
const result = await docushell.parse({ file: document, output: ["json", "markdown"], includeSourceMap: true, failClosed: true});202 acceptedsource map attachedEthos grounded
JSON + MarkdownUse structured fields and readable artifacts from the same parse.
Page + coordinatesOpen the source location associated with an extracted value.
Webhook-ready jobsMove larger document workflows through queued, observable processing.
FREQUENTLY ASKED QUESTIONS
Answers before you integrate.
Direct answers about document inputs, outputs, evidence, and temporary processing.
01What does the DocuShell Parse API return?+
A parse job can return deterministic JSON together with Markdown, HTML, and text artifacts. The structured result can include headings, tables, fields, page references, coordinates, source maps, and explicit partial or warning states.
02How is source-linked evidence different from a citation added by an LLM?+
A generated citation is still a model claim. DocuShell carries document page and region information through parsing so your application can open the exact source context and verify the claim before releasing it.
03Does the Parse API work with scanned PDFs?+
Yes. DocuShell supports born-digital PDFs and OCR-capable scanned-document workflows. When source quality limits extraction, the result can surface a partial or review-required state instead of silently guessing.
04How long are uploaded documents retained?+
Uploads and generated artifacts are temporary. Delivered files are removed after streaming, and remaining processing files are swept within the one-hour retention window.
05Can I evaluate the output before integrating?+
Yes. Use the Parse Playground and sample documents to inspect structured output and source references before creating an authenticated API workflow.
START WITH A REAL DOCUMENT
Give your AI a better understanding of every page.
Parse structure, preserve source context, and let your team verify the evidence before a workflow moves forward.