How DocIntel compares
Honest, feature-level comparison against every major document intelligence tool.
The document AI landscape includes cloud APIs, open-source libraries, and SaaS platforms — each with different strengths. Purple8 DocIntel is a self-hosted, LLM-powered extraction engine that combines format parsing, domain ontologies, and knowledge-graph–ready output in a single container. Below we compare it head-to-head against the tools you're most likely evaluating.
Feature matrix
| Capability | Purple8 DocIntel | Unstructured.io | LlamaIndex | Azure AI Doc Intel | AWS Textract | Google Doc AI | Reducto |
|---|---|---|---|---|---|---|---|
| Deployment model | Self-hosted (Docker / K8s) | Cloud API + OSS | Library (Python) | Azure cloud | AWS cloud | GCP cloud | Cloud API |
| Air-gapped / on-prem | ✔ | ◐ OSS only | ✔ | ✘ | ✘ | ✘ | ✘ |
| BYO LLM (any provider) | ✔ OpenAI / Azure / Anthropic / Gemini / Ollama | ✘ | ✔ | ✘ Azure only | ✘ | ✘ | ✘ |
| Document formats | 70+ (Office, PDF, CAD, BIM, images, code, SAP) | 50+ | ◐ via connectors | ~10 | ~5 (PDF, PNG, JPEG, TIFF) | ~8 | ~10 |
Native MCP Server (/mcp) | ✔ Streamable HTTP (X-API-Key + Sandbox) | ✘ | ✘ | ✘ | ✘ | ✘ | ✘ |
Local Zero-Shot NER (GLiNER v4.1) | ✔ 3,276 docs/hr on CPU (0 LLM calls) | ✘ | ✘ | ✘ | ✘ | ✘ | ✘ |
| CAD / BIM / IFC support | ✔ IFC, DXF, DWG, STEP, STL, GLTF, G-code | ✘ | ✘ | ✘ | ✘ | ✘ | ✘ |
| Entity & relationship extraction | ✔ Two-pass LLM (entities → relationships) | ✘ Chunking only | ◐ via custom code | ◐ Key-value pairs | ◐ Forms / tables | ◐ Entity extraction | ◐ Tables / key-values |
| Domain extraction profiles | ✔ 5 built-in (AEC, Legal, Finance, Scientific, General) | ✘ | ✘ | ◐ Custom models | ✘ | ◐ Specialized processors | ✘ |
| Confidence calibration | ✔ Temperature-scaled sigmoid | ✘ | ✘ | ◐ Raw confidence | ◐ Raw confidence | ◐ Raw confidence | ✘ |
| Knowledge graph output | ✔ Typed entities + relationships (JSON / Graph) | ✘ | ◐ via KG index | ✘ | ✘ | ✘ | ✘ |
| Table extraction | ✔ Dedicated window injection | ✔ | ◐ | ✔ | ✔ | ✔ | ✔ |
| OCR (multiple engines) | ✔ Tesseract / Azure / Google / AWS | ✔ | ◐ | ✔ Azure only | ✔ AWS only | ✔ Google only | ✔ |
| Multi-language support | ✔ 75+ languages (Lingua + prompts) | ✘ | ✘ | ✔ | ◐ | ✔ | ◐ |
| Incremental / dedup processing | ✔ SHA-256 content hashing | ✘ | ✘ | ✘ | ✘ | ✘ | ✘ |
| SSE streaming | ✔ Real-time partial results | ✘ | ✘ | ✘ | ✘ | ✘ | ✘ |
| Source connectors | ✔ SharePoint / Confluence / S3 / Webhooks | ✔ Many connectors | ✔ Many loaders | ✘ | ✘ | ✘ | ✘ |
| Pricing model | License-based (Community Free / Commercial / Enterprise) | Free OSS / Pay-per-page API | Free (MIT) | Pay-per-page | Pay-per-page | Pay-per-page | Pay-per-page |
✔ = Full support ◐ = Partial / limited ✘ = Not available
Head-to-head comparisons
vs Unstructured.io
Unstructured excels at parsing documents into raw chunks. DocIntel goes further — it extracts structured entities and relationships from those chunks using domain-aware LLM prompts.
- Unstructured gives you chunks of text with element types (Title, NarrativeText, Table).
- DocIntel gives you typed entities (Person, Contract, Building), relationships (signed_by, located_in), and confidence scores — ready for a knowledge graph.
- Unstructured has no domain profiles, no confidence calibration, no relationship extraction, no BIM/CAD support.
- DocIntel includes Unstructured-level parsing plus the intelligence layer on top.
vs LlamaIndex / LangChain
LlamaIndex and LangChain are orchestration frameworks — they provide document loaders and LLM chains but leave extraction logic to you.
- LlamaIndex loads documents and builds vector indexes for RAG. It doesn't extract structured entities or relationships out of the box.
- DocIntel is a turnkey extraction API — upload a file, get structured entities + relationships back. No Python code, no prompt engineering needed.
- LlamaIndex requires you to build and maintain extraction chains, prompts, and output parsers yourself.
- DocIntel ships with 5 domain ontologies, confidence calibration, and fuzzy deduplication built-in.
vs Azure AI Document Intelligence
Azure AI Document Intelligence (formerly Form Recognizer) is excellent for forms, receipts, and invoices. DocIntel targets a wider scope.
- Azure Doc Intel extracts key-value pairs from known document types (invoices, receipts, ID cards). Great for forms.
- DocIntel handles arbitrary documents — contracts, research papers, BIM models, CAD files — with domain-aware LLM extraction.
- Azure Doc Intel requires Azure cloud; DocIntel runs in your own infrastructure.
- Azure charges per page ($1–$15 per 1,000 pages). DocIntel is free during beta with flat-rate paid plans coming soon — BYO LLM costs only.
- DocIntel supports 70+ formats including IFC, DXF, DWG, STEP, STL, GLTF, G-code, IDoc, and ABAP which Azure doesn't handle.
vs AWS Textract
Textract is an OCR-first service focused on extracting text, forms, and tables from scanned documents.
- Textract excels at raw OCR, form extraction, and table detection from images and PDFs.
- DocIntel uses OCR as one stage in a deeper pipeline — after parsing, it runs LLM-powered entity and relationship extraction with domain ontologies.
- Textract outputs raw text + bounding boxes. DocIntel outputs typed knowledge-graph–ready entities.
- Textract is AWS-only. DocIntel is self-hosted with pluggable OCR (including Textract as a backend).
- Textract has no concept of extraction profiles, confidence calibration, or incremental processing.
vs Google Document AI
Google Doc AI offers specialized processors for invoices, contracts, procurement, and custom entity extraction.
- Google Doc AI provides pre-built processors and allows custom training. Strong for structured form types.
- DocIntel requires no training data — domain extraction works out of the box via LLM prompts and built-in ontologies.
- Google Doc AI is GCP-only with per-page pricing. DocIntel is self-hosted.
- Google Doc AI doesn't support CAD/BIM formats, doesn't produce relationship graphs, and doesn't offer SSE streaming.
- DocIntel's BYO LLM approach means you're not locked into Google's models or pricing.
vs Reducto
Reducto focuses on high-quality document parsing and structured extraction via their cloud API.
- Reducto provides excellent PDF-to-structured-data conversion with their proprietary models. Cloud-only.
- DocIntel is fully self-hosted — your documents never leave your infrastructure. Critical for regulated industries.
- Reducto charges per page. DocIntel is flat-rate with BYO LLM costs.
- Reducto doesn't support CAD/BIM, doesn't have domain extraction profiles, and doesn't output typed knowledge graphs.
- DocIntel offers pluggable LLM providers — switch from OpenAI to Anthropic to local Ollama without code changes.
Why teams choose DocIntel
🏗️ You work with CAD / BIM / engineering documents
No other tool extracts structured intelligence from IFC, DXF, DWG, STEP, STL, GLTF, and G-code. If you're in AEC, manufacturing, or infrastructure — DocIntel is the only option with native BIM support and a dedicated AEC extraction profile.
🔒 You need self-hosted / air-gapped deployment
Cloud APIs mean your documents leave your infrastructure. DocIntel runs in a single Docker container with optional Ollama for fully local LLM inference. Zero external API calls. Your documents stay yours.
🧠 You need more than raw text
Parsing libraries (Unstructured, Tika, Textract) give you chunks. DocIntel gives you typed entities, relationships, and confidence scores — structured output ready for a knowledge graph, search index, or downstream AI pipeline.
🎯 You need domain-aware extraction
Generic LLM extraction misses domain-specific entities. DocIntel ships with 5 ontology-constrained profiles (AEC, Legal, Finance, Scientific, General) that know the entity and relationship types for each domain. Auto-selected based on document content.
💰 You want predictable costs
Cloud APIs charge per page and costs scale linearly. DocIntel offers a free Community edition and flat-rate paid plans (pricing coming soon) with BYO LLM — no per-page billing surprises.
Choosing the right tool
| If your need is… | Best fit |
|---|---|
| Raw OCR from scanned images | AWS Textract or Google Vision |
| Invoice / receipt field extraction | Azure AI Document Intelligence |
| Chunking documents for RAG | Unstructured.io (OSS) or LlamaIndex |
| Structured entity + relationship extraction | Purple8 DocIntel |
| CAD / BIM / engineering document intelligence | Purple8 DocIntel |
| Self-hosted, air-gapped, BYO LLM | Purple8 DocIntel |
| Knowledge graph–ready output | Purple8 DocIntel |
| Domain-aware extraction (legal, finance, AEC, scientific) | Purple8 DocIntel |
Ready to try DocIntel?
Start with the free Beta licence — 500 documents/month, all features, no credit card.
Get Started →Quickstart →