# Unstract

> LLM-powered platform that turns unstructured documents into structured JSON via prompts, deployable as an API or ETL pipeline.

Unstract, developed by Zipstack, uses LLMs to extract structured JSON from documents such as PDFs, images, and scans. Users define what to extract with natural language prompts and deploy the result as an API or ETL pipeline. The core platform is open source under AGPL-3.0, with managed cloud and enterprise editions also offered.

## What It Is

Unstract is a document extraction platform aimed at teams in finance, insurance, healthcare, and KYC/compliance. Instead of writing regex or per-vendor templates, users write prompts in Prompt Studio that handle layout variations, and the output is JSON ready for a database. The site describes it as document agnostic, working without prior training or templates.

## How the Workflow Runs

Prompts are authored in Prompt Studio, then deployed as a REST API (send a document, receive JSON) or as an ETL pipeline that pulls documents from a source, processes them, and loads results into a warehouse. An MCP server connects the platform to AI agents, and an n8n node lets it drop into existing automation workflows. Supported inputs include PDF, Office documents, spreadsheets, presentations, and common image formats.

## Providers and Connectors

The README lists adapters for LLM providers (OpenAI, Azure OpenAI, Anthropic, AWS Bedrock, Google Gemini, Ollama, Mistral AI and others), vector databases (Qdrant, Pinecone, Weaviate, Milvus, PostgreSQL), and text extractors (LLMWhisperer, Unstructured.io, LlamaIndex Parse). ETL sources include S3, Google Cloud Storage, Azure Blob, Google Drive, Dropbox, and SFTP; destinations include Snowflake, Redshift, BigQuery, PostgreSQL, MySQL, and SQL Server.

## Broader Platform and Editions

The website presents a layered stack: LLMWhisperer for document preparation, Document Studio for structured extraction, and Flow for agentic orchestration with human-in-the-loop handoffs. Cloud and enterprise features listed in the README include LLMChallenge dual-LLM verification, SinglePass and summarized extraction, human review with document highlighting, SSO and RBAC, and compliance certifications (SOC 2, HIPAA, ISO 27001, GDPR).

## Setup Path

The open-source edition runs locally with Docker Compose via a run-platform script on Linux or macOS, with 8 GB RAM minimum.

## Features
- Prompt Studio for natural-language extraction schemas
- REST API deployment returning JSON
- ETL pipelines from storage to data warehouses
- MCP server for AI agents
- n8n custom node
- LLMChallenge dual-LLM verification
- SinglePass and summarized extraction
- Human-in-the-loop review
- SSO and enterprise RBAC
- Multiple LLM, vector database and text extractor adapters
- Self-hosting via Docker Compose

## Integrations
OpenAI, Azure OpenAI, Anthropic Claude, AWS Bedrock, Google Gemini, Ollama, Mistral AI, Qdrant, Pinecone, Weaviate, Milvus, PostgreSQL, LLMWhisperer, Unstructured.io, LlamaIndex Parse, AWS S3, Google Cloud Storage, Azure Blob, Google Drive, Dropbox, SFTP, Snowflake, Amazon Redshift, Google BigQuery, MySQL, SQL Server, Oracle, n8n

## Platforms
MACOS, LINUX, WEB, API, CLI

## Pricing
Open Source, Free tier available

## Version
v0.203.0

## Links
- Website: https://unstract.com
- Documentation: https://docs.unstract.com/
- Repository: https://github.com/Zipstack/unstract
- EveryDev.ai: https://www.everydev.ai/tools/unstract