Build your own document intelligence pipeline

A self-serve API for developers to parse, extract, evaluate, correct, and deploy document AI applications.

IDP Forge

Sort Documents into Your Own Classes and Route Each to the Right Extraction Schema

Classify decides what each document is against a taxonomy you define, with a confidence score at every level. It then uses that answer to pick the extraction schema. Just state your classes and publish.

Classify Tells Your Pipeline What Every Document Is

Mixed inbound streams need sorting before anything else can happen. Hard-coded rules break easily, and a flat list of labels does not match how businesses actually group documents. Classify determines what a document is, such as an invoice, bank statement, or KYC form, based on a class hierarchy you define. It then uses that answer to choose the extraction schema, so classification and routing become one decision instead of a separate classifier plus a hand-coded router.

CORE CAPABILITIES / CLASSIFY

Everything You Need to Sort and Route Documents Without Training a Model

Classify and Split have separate switches, so you can classify only, split only, or do both.

Hierarchical Taxonomies

Build a tree of classes with stable codes, descriptions and classification instructions.

Classify turns mixed document streams into sorted, routed work, so every document reaches the right schema

Insurance Claims Intake

A claims mailbox receives first notice of loss forms, police reports, repair estimates, medical bills and correspondence, all mixed together. Classify sorts each one into the carrier's own class tree and routes it to the right schema.

Routing Documents to the Right Extraction Schema

The class a document lands in decides which schema extracts it, so invoices, bank statements and KYC forms each get the extraction built for them

Sorting Mixed Inbound Streams

Mailroom batches, shared intake folders and watched Google Drive folders deliver every document type at once. Classify labels each document as it arrives, so nothing waits for manual sorting

Classifying After Split

Classifying a whole package gives a meaningless answer. When Split cuts a bundle into documents, Classify labels each child on its own and sends it to its own schema

Labeling and Organizing Document Archives

Use a label taxonomy to tag documents by type without extracting anything, so back files and archives become searchable and organized.

Handling Unknown and Uncertain Documents

Unrecognized documents land in Others and are still captured. Uncertain ones wait for a reviewer to confirm or change the class, so nothing is dropped or sent to the wrong schema.

Pick the Right Engine for Every Sub-Stage

Each sub-stage in Parse has its own on/off switch, engine, model and optional confidence threshold. Use the engine that is strongest for each one, and change it later without writing code.

01

Choose the model for each sub-stage. The picker filters by Vision, Reasoning and Fast, and shows a relative cost tier from $ to $$$.

Layout
Vision Fast
$$
OCR
Fast Reasoning
$$$
Chunking
Fast Vision
$
02

Bring your own LLM keys. Use your own keys for the LLM-based sub-stages. Azure Document Intelligence usage is always billed in credits.

OpenAI key connected LLM stage
Claude key connected LLM stage
Azure Doc Intelligence credits
Billing note: Azure Document Intelligence is always billed in credits.
03

Built to recover. When a model call fails, the stage retries automatically (3 attempts with backoff) and rotates provider keys and deployments.

Model call failed
retry trigger
Try 1 Try 2 Try 3
Provider key rotated · deployment switched
auto recover
Stage completed successfully
recovered

CLASSIFY / PROCESS FLOW

How Classify Works

Built to Plug into Your Stack

Submit a file to a parse-only pipeline with one call:

curl -X POST \
  https://api.us.idpforge.cloud/v1/jobs \
  -H "X-API-Key: $IDP_API_KEY" \
  -F "files=@policy-handbook.pdf" \
  -F "pipeline_id=$PARSE_PIPELINE_ID"

Frequently Asked Questions

Does it handle scanned and handwritten documents?

Scans and photos, yes. OCR runs on page images. Handwriting quality depends on the engine you choose (Azure Document Intelligence or a vision LLM). There is no separate handwriting mode.

‍

Do I get Markdown or JSON?

Both. You get JSON with whole-document Markdown, per-page Markdown and typed blocks. Post Processing can also render CSV.

Are tables preserved?

Yes. Tables are detected as table regions and output as Markdown tables.

Can I get bounding boxes?

Yes. Every block has a bounding box, expressed as percentages of the page.

Can I chunk for my vector database?

Yes. Choose semantic, page or fixed-token chunking, with configurable size and overlap.

Which OCR engine is used?

Azure Document Intelligence by default. You can select a vision LLM per pipeline.

Is there a page limit?

There is no hard page limit. The file size limit is 100 MB.

Can I use my own LLM key?

Yes, for the LLM-based sub-stages. Azure Document Intelligence usage is billed in credits.

How long is my data kept?

Jobs, results and documents are kept for 24 hours after submission, then deleted.

Build your first document AI workflow in minutes.