Skip to Content
ProcessorsOverview

Processors

Processors are the intelligence layer of M3 Forge, transforming raw documents into structured, actionable data. They power document understanding through classification, extraction, splitting, layout analysis, and summarization using Generative AI and machine learning models you can train on your own data.

Why Processors Matter

Modern document processing requires more than simple parsing. Processors provide:

  • Data Extraction — Pull structured fields from unstructured documents
  • Document Classification — Identify document types automatically (invoices, contracts, forms, etc.)
  • Smart Splitting — Break multi-document files into logical segments
  • Layout Analysis — Understand document structure and vendor-specific format variations
  • Summarization — Generate concise summaries for short and long documents

All processors are powered by Generative AI. Custom processors support training on your own data, so models learn from your specific document types and business logic — with as few as 10 examples.

Processor Types

M3 Forge supports five custom processor types:

TypeDescriptionUse Cases
Custom ExtractorIdentify and extract specific data from your documentsInvoice line items, contract terms, form values, table data
Custom ClassifierGroup your documents into categoriesDocument routing, type detection, topic classification
Custom SplitterIdentify document boundaries in a large fileBatch processing, scan separation, email attachments
Custom LayoutClassify documents by layout variation, such as vendor-specific formatsMulti-vendor invoices, form templates, layout-dependent extraction
SummarizerGenerate summaries for short and long documentsDocument digests, executive summaries, content previews
Custom Processors page showing the five processor type cards with Create processor buttons

All processor types are tagged as Generative AI and integrate seamlessly with workflows. Use processors as workflow nodes to build end-to-end document intelligence pipelines.

The Processors section in the sidebar provides three views:

PagePathPurpose
Custom Processors/processors/customCreate new custom processors from the five type templates
Processor Gallery/processors/galleryBrowse pre-built processors organized by category
My Processors/processors/my-processorsView and manage your created processors

Processing Pipeline

A typical document processing pipeline combines multiple processor types:

  1. Classification — Identify document type
  2. Splitting — Separate individual documents from batch files
  3. Layout Analysis — Understand structure and vendor-specific formats
  4. Extraction — Pull structured fields
  5. Summarization — Generate document summaries
  6. Validation — Human-in-the-loop review if needed

Processors handle the heavy lifting. Workflows orchestrate the pipeline.

Training Custom Processors

All custom processor types support training on your data:

  1. Create Processor — Select type and name your processor
  2. Import Training Data — Upload example documents
  3. Label Examples — Annotate documents with ground truth
  4. Train Model — Launch training jobs with your labeled data
  5. Evaluate Performance — Review metrics and confusion matrices
  6. Deploy — Activate trained models for production use

Training happens in M3 Forge’s labeling interfaces. No code required.

Training data quality matters more than quantity. Start with as few as 10 representative examples and iterate. Diverse, well-labeled examples produce better models than thousands of similar documents.

Pre-Built Processors

Beyond custom processors, M3 Forge includes pre-built processors for common document types:

CategoryDescriptionExamples
GeneralReady-to-use processors for common document tasksDocument OCR, Form Parser, Layout Parser
SpecializedDomain-specific schematized processorsInvoice Parser, Receipt Parser, Bank Statement Parser, Pay Slip Parser, Identity Document Proofing, US Driver License Parser, US Passport Parser, Utility Parser

Browse all available processors in the Processor Gallery.

Processor Lifecycle

Processors follow a versioned lifecycle:

  • Draft — Initial configuration, not yet trained
  • Training — Model training in progress
  • Ready — Trained and ready to deploy
  • Active — Currently deployed in production
  • Archived — Retired from active use

Each training run creates a new version. Roll back to previous versions if needed.

Next Steps

Explore processor capabilities:

Custom Processors

Create and train processors on your own documents

Processor Gallery

Browse pre-built processors and deploy ready-to-use models

Annotators

Configure labeling interfaces for training data

Training

Launch training jobs and evaluate model performance

Integration Points

Processors integrate with:

  • Workflows — Use processors as workflow nodes for automated pipelines
  • HITL — Route low-confidence predictions to human review
  • Knowledge Base — Extract entities for semantic search indexing
  • Agents — Provide structured data for AI agent reasoning

Learn more about building document intelligence pipelines in the Workflows section.

Last updated on
SearchCustom Processors