Unstract LLM-Powered Document Extraction Platform

Unstract is the document AI brand of Zipstack, Inc., whose contact address is in Los Altos, California. It targets the gap between experimenting with large language models on documents and running reliable extraction in production for invoices, bank statements, purchase orders, bills of lading, insurance packets and legal files.

The platform has two layers. LLMWhisperer prepares documents for LLMs by producing layout-preserved text from scanned PDFs, phone photos, forms and more than 20 file types, with checkbox detection and a Highlight API that links answers to their place on the page. Document Studio and Prompt Studio let teams define a JSON schema and prompts once and map varied layouts to it, with an agentic mode that drafts schemas and validates accuracy. Extractions deploy as APIs, ETL pipelines into warehouses, or MCP servers for AI agents, and the newer Flow product orchestrates multi-step processes.

• LLMWhisperer: layout-preserving text extraction from scans, photos and forms for LLM input.
• Prompt Studio: define schemas and prompts once and reuse them across document variants.
• Agentic Prompt Studio: drafts schemas, crafts prompts and validates extraction accuracy.
• Deployment modes: publish extractions as API, ETL pipeline or MCP server.
• Ready APIs: invoice, bank statement, purchase order, bill of lading and table extraction.
• Human in the loop and LLMChallenge: reviewer checks and cross-model verification of outputs.

Vietnamese banks can self-host the open-source edition to extract bank statement and loan document data without sending files abroad. Freight forwarders can turn bills of lading and packing lists into ERP records. Software vendors can embed LLMWhisperer as the parsing layer for AI assistants that answer questions over customer documents.

Teams can choose their own LLM and keep data on premises or in their own AWS, Azure or GCP VPC. LLMWhisperer reads more than 300 printed languages and handwriting in 12, and a free tier allows 100 pages a day without a credit card. The company reports 600+ paying LLMWhisperer customers and more than 10 million pages processed per month in Document Studio.

Enterprises are replacing template-based IDP with LLM-native extraction, but most pilots stall before production because of accuracy and data control concerns. Open-source and self-hosted options appeal to Southeast Asian banks, insurers and logistics firms that must keep customer documents inside their own infrastructure.

Zipstack, Inc. does business as Unstract and lists its contact address at 23175 Eastbrook Avenue, Los Altos, California. The platform is available as managed cloud, on-premise and open-source editions, with code on GitHub (Zipstack/unstract). The website shows GDPR, ISO, SOC and HIPAA compliance badges, and LLMWhisperer holds a 4.6 out of 5 G2 rating as a Category Leader.