Agnos
Download

Agnos vs Unstructured

Last updated

Choose Unstructured if you write code, need scanned pages read with OCR, or run document pipelines on a Mac, on Linux, on servers or in the cloud. Choose Agnos if you want a Windows app that cleans text-based documents on your own PC with no code, and shows what it removed and the rule for each.

Gidion makes Agnos and wrote this page. Every claim about Unstructured comes from Unstructured’s own website, documentation and GitHub, checked on 17 October 2026, and each source is linked under Sources. Unstructured offers a free, open-source Python library and hosted products, Transform and Pipelines. It calls the library a starting point for prototyping and recommends Pipelines for production.

At a glance

AgnosUnstructured
What it isA desktop app, made by Gidion, that cleans documents before they go into AI tools, using named rules.An open-source Python library that splits documents into typed elements, plus hosted products: Transform (an API, a web app and an MCP server) and Pipelines (a no-code platform for document pipelines).
Who it’s forPeople preparing files for Custom GPTs, NotebookLM, Claude Projects, Copilot, a RAG index or training data, without writing code.Developers building document pipelines for language models (the library), people sending documents from an AI tool or a web app (Transform), and companies processing documents at scale (Pipelines).
Where it runsOn your Windows PC. Documents are never uploaded; Agnos goes online only to activate, check and deactivate a licence.The library: on your own computer, server or container. Transform: on Unstructured’s cloud. Pipelines: on Unstructured’s cloud or, on the Business plan, in a dedicated instance, in your own AWS, Azure or GCP account, or on bare metal.
Needs codeNo.The library: yes, Python. Pipelines: no, it has a no-code interface as well as an API. Transform: no for its web app or for plain-language prompts in an AI tool; its API and SDKs are code.
File typesText-based PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx), web pages, CSV and TSV, JSON and JSONL, Markdown, plain text and source code. Older Office, RTF and OpenDocument files must be saved as .docx, .xlsx or .pptx first.The library: PDF, Word (.doc, .docx), PowerPoint (.ppt, .pptx), Excel (.xls, .xlsx), HTML, Markdown, text, CSV, TSV, email (.eml, .msg), EPUB, RTF, OpenDocument text, XML and images. Pipelines and the Transform MCP server take more; the Transform web app and API take PDF, .docx, .pptx and images.
Scanned PDFs (OCR)Not read. A page with no text layer is skipped and named in the file’s warnings, and a PDF with no text at all is refused.Yes. The library uses Tesseract, installed separately, when a PDF has no text of its own. The hosted products add vision language models and generative OCR.
TablesRead from Word, PowerPoint and web pages, and from PDFs with the slower Layout reading mode. Excel sheets become tab-separated rows. The RAG purpose keeps tables exactly as written. Only the Sections split keeps a table whole.Tables can come out as Table elements with an HTML copy of the table. In the library, the hi_res strategy is the one that extracts tables from PDFs. The hosted products add table descriptions and table-to-HTML.
What it outputsA RAG corpus in JSONL, with each piece’s pages or headings; cleaned .txt files, one per document; training data in JSONL, with an optional validation split; or Everything: one zip with the dataset, every removed piece and why (rejected.jsonl), a manifest and a dataset card.Typed elements, such as Title, NarrativeText, Table, Header and Footer, with metadata. The library saves them as JSON, CSV, Markdown or text, and can chunk them. The Transform API returns the parsed text as Markdown, in a JSON result. The Transform MCP server returns Markdown, JSON, HTML or text, with optional chunks and embeddings. Pipelines writes JSON to more than 20 destinations, including vector databases.
Shows what it removed, and whyYes. Review lists what it removed and the rule for each, and the Everything export saves the full list as rejected.jsonl. If cleaning makes more than 500 removals in one file, Review counts them all but lists the first 500.It labels elements by type, including Header, Footer and PageNumber. Its docs say you can filter out the types you don’t need, in your own code.
PriceFree for 14 days, with everything in Pro and no card. Then Pro, $99 once (one person, up to 2 devices), or the Company Pack, $399 once (up to five people at one company, up to 10 devices). Every 2.x update included. Tax is added where it applies. No subscription.The library is free. The hosted products: 10,000 free pages to start, with no card, then $0.015 a page on Pay-As-You-Go. Business has custom pricing.
Operating systems64-bit Windows 10 (version 2004 or later) and Windows 11, from the Microsoft Store.The library: macOS, Linux and Windows, and Docker images for x86_64 and Apple silicon. The hosted products are used from a browser, an AI tool or your own code.
LicenceNot open source. Free for the 14-day trial; after it, exporting needs a licence, bought once, under our terms.The library: Apache License 2.0. The hosted products are commercial services with a free start; Business needs a signed agreement.

The Agnos column describes Agnos 2.3.2. The Unstructured column describes its website and docs on 17 October 2026, when the library’s latest release was 0.27.28, published on 16 October 2026.

Where Unstructured is the better choice

  • Scanned documents and images. Unstructured reads text out of scans and pictures. Its library runs Tesseract OCR when a PDF has no text of its own, and the hosted products add vision language models and generative OCR. Its home page says it captures text from scans, images, handwriting and low-quality documents. Agnos does not read text out of pictures at all.
  • Tables and complex layouts. The library can detect a page’s layout, and keeps a table as an element with an HTML copy of its rows and columns. The hosted products add table descriptions and reading order detection. Unstructured publishes its own benchmark on more than 1,000 enterprise pages, including scans and nested tables, against other parsers. Its best results come from its hosted pipelines, and its open-source library is listed separately with lower scores. These are Unstructured’s results, and we have not tested them. Agnos reads tables only as text, and in PDFs only with its slower Layout mode.
  • More file types. The library reads email (.eml, .msg), EPUB, older Office files (.doc, .xls, .ppt), RTF, OpenDocument text and images, and the hosted products take dozens of formats. Agnos does not read email, EPUB or images, and asks you to re-save older Office, RTF and OpenDocument files as .docx, .xlsx or .pptx.
  • Mac, Linux, servers and the cloud. The library installs on macOS, Linux and Windows and comes as a Docker image for x86_64 and Apple silicon. The hosted products run on Unstructured’s cloud, or for Business customers in a dedicated instance, in their own AWS, Azure or GCP account, or on bare metal. Agnos is a Windows desktop app only.
  • Pipelines and scale. Unstructured Pipelines connects to more than 20 sources, such as S3, SharePoint and Google Drive, and more than 20 destinations, including vector databases such as Pinecone, Qdrant and Weaviate. It chunks, adds embeddings and runs on a schedule. Unstructured’s free, open-source Ingest library and command line also process files in batches and connect to sources and destinations, though Unstructured recommends its API instead. Agnos makes files that you load yourself.
  • Inside your AI tools. The Transform MCP server works from Claude Desktop, Claude Code, Cursor, Visual Studio Code and many other MCP clients, and Transform can extract fields from a document into a JSON schema you define.
  • Open source, and free to start. The library is under the Apache License 2.0, so you can read and change the code, and Unstructured says it will stay free. The hosted products give you 10,000 pages free to start, with no card.
  • Company requirements. Unstructured’s pricing page lists SOC 2 Type 2, ISO 27001, HIPAA, GDPR and role-based access for its hosted products, and multi-user accounts on the Business plan. Its home page adds FedRAMP High.

Where Agnos is the better choice

  • No code and no setup. Install Agnos from the Microsoft Store and drop a folder on it. Unstructured’s library is Python, and for some file types it also needs Poppler, Tesseract and LibreOffice installed. Unstructured’s hosted products need no code either, but they process your files on Unstructured’s servers or, on the Business plan, in your company’s own cloud or on bare metal.
  • Nothing is uploaded. Agnos reads, cleans and exports on your PC. It goes online only for its licence, and sends no document, file name or text; it has no analytics. Transform keeps uploaded files in temporary storage for 24 hours, and Unstructured’s pricing page lists encryption in transit and zero data retention among its security features. Unstructured’s library also runs on your own machine; its README says it sends small usage pings by default, never with document content, and explains how to turn them off.
  • It shows what it removed, and why. Review lists what it removed and the rule for each piece, such as “A running header, footer or page number at the edge of a page”. Lines like that one are removed before the file is cut into pieces, so they come back when you switch their rule off for the project and clean again. Other removed pieces can be put back one at a time with Keep this anyway. The Everything export saves the full list. Unstructured labels headers, footers and page numbers as element types, and its docs say you can drop them in your own code.
  • Rules made for clutter. Named rules remove cookie banners, “enable JavaScript” notices, website menus, running headers and footers, page numbers, placeholder text, and passages that repeat, within a file when cleaning and across files when you export. Unstructured’s library offers cleaning functions that you call on text in your code, such as removing bullets, dashes and extra spaces.
  • One price, paid once. Pro is $99 once and the Company Pack is $399 once, with every 2.x update included and no charge by the page. If your pages need OCR, or you only have a few, Unstructured’s free library or its 10,000 free pages may suit you better.
  • Mixed folders, ready for AI tools. Drop a folder of PDFs, Word, Excel and PowerPoint files and saved web pages, subfolders included, and each file is cleaned as it arrives. Export cleaned .txt files for Custom GPTs, NotebookLM, Claude Projects or Copilot, a RAG corpus in JSONL, or training data with a validation split.

Using them together

Agnos cannot read scans, email or EPUB books, and Unstructured can. Agnos reads Markdown and plain text files, so Unstructured’s output can go into Agnos:

  1. Read the files with Unstructured. Use the library: for a PDF with no text of its own, its default auto strategy chooses ocr_only, which reads the page with Tesseract, and hi_res also detects the layout and extracts tables. Or send the files to the Transform MCP server.

  2. Save the result as Markdown or text. The library has elements_to_md and elements_to_text, and the Transform MCP server can return md or txt. If you like, drop the elements typed Header, Footer and PageNumber first.

  3. Add the files to an Agnos project with your other documents. Agnos cleans them with the same rules, lists what it removed in Review, and exports them with the rest.

Two things to know. A text file has no pages, so Agnos’s rule for the edges of PDF pages has nothing to look at; it still removes page markers such as “Page 3 of 10” on lines of their own. And if you use Transform, your files go to Unstructured’s servers.

Which to try

If your documents are scans, or you need a pipeline on Linux or in the cloud, start with Unstructured. If they are text-based files on a Windows PC and you want to see exactly what was taken out before they go into an AI tool, try Agnos on a folder of your own: everything in Pro is free for 14 days, with no card. Get it from the download page.

Sources

Unstructured’s own pages, checked 17 October 2026:

Agnos’s own documentation: Supported files, Cleaning settings, Export, Export formats, Review, Your files and privacy, What Agnos doesn’t do, Pricing and the licence terms.

Unstructured is a trademark of its owner. This page is ours, not theirs; tell us at hello@agnos.cloud if anything here is out of date.