Agentic Document Extraction
LandingAI
Parse and extract structured data from documents, with citations
Agentic Document Extraction (ADE) is LandingAI's document-intelligence platform for turning PDFs, images and Office files into structured, auditable data. Parse converts a document into LLM-ready Markdown with layout-aware structure and per-block grounding (page number, coordinates and table-cell citations); Extract pulls specific fields against a JSON schema you define, including nested objects, arrays and large multi-page tables; Split segments multi-document files into classified sub-documents. It is delivered as modular REST APIs with Python and TypeScript libraries, an ade CLI, an MCP server and a browser Playground, with async jobs and signed webhooks for large batches.
Primary Functions
Features
- Parse produces LLM-ready Markdown with layout-aware structure and structured content blocks - text, tables and figures - with hierarchy preserved
- Precise citations for every block (page, coordinates and table-cell grounding) plus confidence scoring that surfaces results which may need human review
- Schema-first Extract supporting flat or nested schemas, arrays and multi-table output, large tables spanning thousands of rows across many pages, and bounding-box citations per value
- Split segments long multi-hundred-page files into classified sub-documents, with instance detection on repeated identifiers such as invoice number, date or order ID
- Async Parse Jobs and Extract Jobs with signed webhooks and independent priority and standard service tiers
- Python and TypeScript libraries, an ade CLI with resumable jobs and stable JSON output for CI, an MCP server, and a no-code Playground
- US and EU hosted regions for data residency, plus VPC and on-premises containerized deployment on Enterprise
Limitations
- Language coverage is uneven: ADE has only 'Basic Support' for Amharic, Bengali, Gujarati, Malayalam, Tamil, Telugu, Urdu and other non-Latin or stylized scripts, where the vendor says it 'may struggle'
- The DPT-3 Verity parsing model reads Latin-script documents only; the published language list applies to DPT-2 and DPT-3 Pro
- Hard request limits: 50 MiB per synchronous document (async jobs allow 1 GiB for PDFs, 50 MiB for images), a maximum of 6,000 pages per PDF, and about 179 megapixels per image
- Zero Data Retention is a Team and Enterprise setting and is not available on the Explore plan; SSO and IP allowlisting are Enterprise-only, and disabling ZDR later requires contacting support
- Rate limits bind early on Explore (100 pages/minute priority, 6,000 pages/hour standard), and by design a document with more pages than the per-minute priority limit can never run on the priority tier - it returns 429 no matter how long you wait