Unstructured

What it does

Unstructured takes in documents such as PDFs, Word and Excel files, email, HTML, images and scans, and produces parsed content, structured elements, tables, chunks and embeddings. Its pipeline covers partitioning, chunking, enrichment, embedding and loading.

People use it to prepare large, messy document collections for AI applications: connecting to sources such as S3, SharePoint or Google Drive, processing files on a schedule, and writing results to vector databases such as Pinecone, Qdrant or Weaviate. Its distinct feature is that it runs the whole ingestion pipeline rather than parsing one file at a time, with table and layout preservation for complex documents.

It suits teams building retrieval or AI workflows over document-heavy data, including finance, legal, insurance and healthcare. Access is through a web app and an API, with a free allowance of 10,000 pages to start and a pay-as-you-go rate of $0.015 per page after that. Limitations include custom pricing for Business and dedicated deployments, and some connectors available only on request.

What it can do

• Parses PDFs, scans, Word, Excel, PowerPoint, email and HTML files through one pipeline
• Extracts tables while preserving rows, columns and headers
• Applies OCR to scanned documents and handwriting
• Preserves reading order and section hierarchy across pages
• Chunks, enriches and embeds parsed content for retrieval
• Connects to 20+ sources and 20+ destinations including major vector databases

Works with

• Amazon Bedrock
• Anthropic Claude
• Azure AI Studio
• IBM watsonx
• Gemini
• OpenAI
• NVIDIA
• Together.ai
• VertexAI
• S3

Example workflows

Parse a single PDF through the API

Illustrative example. Input: a PDF saved as document.pdf and an API key. Steps: send a POST request to the Parse endpoint with the file; read the returned markdown, metadata and source fields. Output: markdown text of the document plus page count and warnings.

Load document chunks into a vector database

Illustrative example. Input: a document source such as an S3 bucket and a configured destination such as Pinecone. Steps: connect the source, run partitioning, chunking and embedding, and write results to the destination on a schedule. Output: embedded document chunks stored in the vector database.

Before you choose

• The 10,000-page allowance is introductory rather than recurring.
• Business and dedicated deployments require a quote; standard pay-as-you-go has a published page rate.
• Some connectors need enablement, and features marked Coming Soon are not treated as available.

Listed under

• Data
• Data extraction
• Data pipeline
• Developers
• Document parsing
• Enterprise
• OCR
• RAG ingestion

Get started

Free allowance of 10,000 pages with no card required, per the pricing page.

1. Sign up at https://transform.unstructured.io/?entry=cta and log in.
2. Obtain an API key and set it as UNSTRUCTURED_API_KEY in your environment.
3. Send a test file to https://transform.unstructured.io/api/v2/parse using the unstructured-api-key header.

New accounts receive a one-time allowance of 10,000 pages without a card. Pay-as-you-go processing then costs $0.015/page. Business, dedicated-instance and VPC deployments have custom terms. Some connectors require enablement; confirm which processing features are generally available.

Source: unstructured.io/pricing, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Convert Documents into Structured, AI-Ready Data | Unstructured ↗unstructured.io
• Unstructured Pricing: Free Tier, Pay-As-You-Go and Enterprise Plans | Unstructured ↗unstructured.io
• Parse your first document - Unstructured ↗docs.unstructured.io

Company website

unstructured.io: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.