Purple8 DocIntelUniversal Document Intelligence (v0.12.5)

Parse 70+ document formats across 75+ languages — PDF, CAD, BIM, Office, images, email, code, SAP — into typed entities, relationships, and structured tables. MCP-native for AI agents. Self-hosted container. Zero external DB dependencies.

AI Agents & MCP Guide →

Docker Quickstart

Machine-Readable (/llms.txt)

🤖

MCP-Native for AI Agents (/mcp)

Connect Claude, Cursor, VS Code Copilot, or custom agents directly to /mcp over Streamable HTTP using X-API-Key or Bearer tokens. Agents call docintel_ingest_url, docintel_ingest_text, and docintel_get_job with zero glue code.

MCP Agent Guide →

📑

70+ Document Formats Natively Parsed

PDF, DOCX, XLSX, PPTX, HTML, TXT, MD, EML, MSG, EPUB, CSV, JSON, XML, RTF, ODT, IFC, IFCZIP, DXF, DWG, STEP, STL, OBJ, GLTF/GLB, G-code, SAP IDoc, ABAP, images, code & config — no pre-conversion step.

All 70+ formats →

🌍

75+ Languages (Lingua + Multi-Language Prompts)

Automatic language detection across 75+ languages paired with language-specific entity naming, transliteration, date, and JSON key normalisation conventions.

Language support →

⚡

353.4 Pages/s & Embedded Durable Job Store

Multi-core CPU worker pool delivers 35.34 docs/s (353.4 pages/s, 70.7 tables/s) parse+layout throughput and 0.91 docs/s (3,276 docs/hr) local GLiNER v4.1 zero-shot NER. Jobs persist in an embedded p8l log — no Redis required.

🏗️

CAD & BIM Intelligence

IFC (2x3 / 4 / 4x3), IFCZIP, DXF, DWG, STEP, STL, OBJ, GLTF, G-code — structured 10-section output covering storeys, spaces, elements, MEP systems, materials, quantities, and property sets.

AEC & BIM →

🧠

Hybrid GLiNER v4.1 + Two-Pass LLM Extraction

Combine local zero-shot GLiNER v4.1 NER with two-pass domain-ontology LLM extraction (entities first, relationships second), fuzzy deduplication (rapidfuzz WRatio ≥ 88), and temperature-scaled confidence calibration.

Confidence calibration →

📊

Structured Table Extraction & Graph Linking

Tables extracted from PDFs, DOCX, and XLSX are preserved as structured rows and columns, injected into dedicated context windows, and queryable in 0.65 ms P50 when committed to Purple8 Hyper Graph.

Table extraction →

🕸️

Standalone JSON or Paired with Hyper Graph

Use DocIntel standalone for structured JSON, SSE streaming, and webhooks — or set GRAPH__BASE_URL to emit entities, relationships, and tables directly into Purple8 Hyper Graph (/ingest/commit).

Compare DocIntel →

How It Works (6-Stage Pipeline)

Stage	What happens
1 · Ingest & Dedup	Upload file (`/process`), submit URL (`/process-url`), or call `docintel_ingest_url` via MCP. SHA-256 content hash returns cached results immediately for unchanged documents.
2 · Detect Format & Language	Extension + magic-byte detection across 70+ formats and Lingua detection across 75+ languages.
3 · Parse & Extract Tables	Format-specific parser builds a canonical `ParsedDocument` while preserving multi-row tables for structured row/column indexing.
4 · Classify Domain	Selects the optimal domain ontology profile (`aec`, `contract`, `financial`, `scientific`, or `general`).
5 · Extract & Calibrate	Local GLiNER v4.1 NER + two-pass LLM entity & relationship extraction, fuzzy deduplication, and temperature-scaled (`T=2.0`) confidence calibration.
6 · Return JSON or Commit to Graph	Returns typed JSON directly (`auto_commit=false`) or commits entities, relationships, and structured tables to Purple8 Hyper Graph (`POST /ingest/commit`).

AI Agent & Container Quickstart (60 Seconds)

For AI Agents reading this page: Fetch https://docs.docintel.purple8.ai/llms.txt or /getting-started/mcp-agents for the complete MCP tool definitions, REST API schema, and dual DocIntel + Hyper Graph configuration.

1. Run Docker Container2. Connect Claude Code / CLI3. Connect Cursor / VS Code / Claude Desktop

bash

docker run -d \
  --name purple8-docintel \
  -p 8200:8200 \
  -e LLM__API_KEY=sk-... \
  -e SERVER__API_KEY=p8d_your_secret_key \
  -v docintel-data:/data \
  ghcr.io/purple8-technologies/purple8-docintel:latest

bash

claude mcp add --transport http purple8-docintel "http://localhost:8200/mcp" \
  --header "X-API-Key: p8d_your_secret_key"

json

{
  "mcpServers": {
    "purple8-docintel": {
      "url": "http://localhost:8200/mcp",
      "headers": {
        "X-API-Key": "p8d_your_secret_key"
      }
    }
  }
}

bash

# Submit a document
curl -X POST http://localhost:8200/process \
  -H "Authorization: Bearer $API_KEY" \
  -F "file=@contract.pdf"

# Stream partial results
curl -N http://localhost:8200/jobs/abc-123/stream \
  -H "Authorization: Bearer $API_KEY"

Full quickstart → · API reference →

Head-to-Head Comparison

Capability	Purple8 DocIntel (`v0.12.5`)	Unstructured.io	Azure AI Doc Intel	AWS Textract	LlamaParse
Deployment	Self-hosted Docker / Air-gapped	Cloud API + OSS	Azure Cloud	AWS Cloud	Cloud API
Document formats	70+ (Office, PDF, CAD/BIM, SAP, Code)	50+	~10	~5	~15
CAD & BIM (`IFC`, `DXF`, `STEP`, `STL`)	✅ Native 10-section parser	❌	❌	❌	❌
Languages supported	75+ languages	Partial	✅	Partial	Partial
Native MCP Server (`/mcp`)	✅ Streamable HTTP (`X-API-Key`)	❌	❌	❌	❌
Local Zero-Shot NER (`GLiNER v4.1`)	✅ Built in (`3,276 docs/hr` CPU)	❌	❌	❌	❌
Two-pass Entity + Relationship Graph Output	✅ Typed nodes, edges & tables	❌ Chunking only	❌ Key-value	❌ Forms/tables	❌ Markdown only
External DB requirement	0 (Embedded `p8l` store)	External	Cloud	Cloud	Cloud

Full competitor comparison → · AI Agents & MCP Guide → · Self-hosting guide →

Purple8 DocIntelUniversal Document Intelligence (v0.12.5)

MCP-Native for AI Agents (/mcp)

70+ Document Formats Natively Parsed

75+ Languages (Lingua + Multi-Language Prompts)

353.4 Pages/s & Embedded Durable Job Store

CAD & BIM Intelligence

Hybrid GLiNER v4.1 + Two-Pass LLM Extraction

Structured Table Extraction & Graph Linking

Standalone JSON or Paired with Hyper Graph

How It Works (6-Stage Pipeline) ​

AI Agent & Container Quickstart (60 Seconds) ​

Head-to-Head Comparison ​

How It Works (6-Stage Pipeline)

AI Agent & Container Quickstart (60 Seconds)

Head-to-Head Comparison