Enterprise document workflows routinely involve files that exceed what a single synchronous request can handle: know-your-customer (KYC) packets spanning 200-300 pages per client, annual reports and regulatory filings in the hundreds of pages, and loan origination packages combining multiple dense documents. LandingAI Agentic Document Extraction (ADE) processes these through an asynchronous parsing path built for large, long-running documents, demonstrated at a global Tier-1 bank that processes multi-lingual corporate documents of 200-300 pages per client with a 40-60% reduction in manual review time, per the Tier-1 bank KYC case study.
Large Documents Run on an Asynchronous Path
Large, multi-hundred-page and multi-thousand-page documents run through ADE's asynchronous parsing path, which decouples submission from retrieval so a long parse never holds open a live connection.
- Submit the document, then retrieve the parsed output when the job completes, the correct pattern for batch pipelines and long-running documents.
- Documents can be submitted from, and results written back to, your own cloud storage, including Amazon S3, Azure Blob Storage, and Google Cloud Storage.
- For the current file-size and page ceilings on a single job, see the asynchronous parsing documentation.
Throughput Scales with Plan
Rate limits apply at the organization level so throughput stays predictable rather than burst-and-block: all requests from any user or key in an organization draw from the same limit.
- Higher plan tiers carry higher rate limits, and Enterprise plans support customizable limits for high-volume pipelines, detailed in the plan and feature breakdown.
- Queue-level monitoring lets pipelines track submitted jobs across an organization.
Zero Data Retention at Scale
When the Zero Data Retention (ZDR) option is enabled, it applies automatically to every asynchronous parse job with no special request parameters, and customer data is not persisted beyond processing across LandingAI and all subprocessors.
For the strictest control over highly sensitive documents, teams can keep document content out of the request and response entirely by submitting documents from, and writing results back to, their own pre-signed cloud storage URLs, so LandingAI does not hold document content at any stage.
Resilient Handling of Partial Failures
If some pages in a large document cannot be parsed, ADE returns the successfully parsed pages and flags the ones that failed rather than failing the entire job.
- Fault-tolerant pipelines check for flagged pages before passing output to the extraction stage.
- Documents with failed pages route to a review or re-submission queue instead of being treated as complete, following the parse troubleshooting guidance.
Batch and Parallel Processing via the Libraries
The Python and TypeScript libraries handle large-document and multi-file workloads without custom orchestration.
- Oversized documents are split, processed in parallel, and stitched back into a single output object.
- A list of files can be submitted in one call, with the library managing parallel processing and automatic retries on transient errors.
FAQ
How large a document can ADE process asynchronously?
ADE's asynchronous parsing path is built for large, multi-hundred-page and multi-thousand-page documents that exceed what a single synchronous request handles. For the current file-size and page ceilings on a single job, see the asynchronous parsing documentation.
Does ZDR work for large documents processed asynchronously?
Yes. When the ZDR option is enabled, it applies automatically to every asynchronous parse job with no special request parameters. For the strictest control over sensitive documents, teams can keep document content out of the request and response by submitting from and writing results back to their own pre-signed cloud storage, including Amazon S3, Azure Blob Storage, and Google Cloud Storage. See the ZDR documentation.
How are rate limits applied when processing large document volumes?
Rate limits apply at the organization level, so all requests from all keys in an organization count against the same limit for predictable throughput. Higher plan tiers carry higher limits and Enterprise plans support customizable limits, detailed in the plan and rate limit documentation.
Can ADE handle documents distributed across multiple files, such as a KYC packet in several PDFs?
For separate files, run each as its own parse job and use the libraries' async clients to process them concurrently. For several documents bundled inside a single PDF, the Split API classifies and separates them into sub-documents, and is supported on the v2 APIs for production use.
Is ADE appropriate for continuous high-volume pipelines processing thousands of pages per hour?
ADE is built for continuous enterprise workloads, with rate limits that scale by plan tier and Enterprise plans supporting customizable limits. The asynchronous path adds queue-level monitoring and the libraries include automatic retry for resilient operation; contact LandingAI through the enterprise contact page to discuss throughput requirements for high-volume continuous workloads.