Agnos
Download

Agnos vs LlamaParse

Last updated

Choose LlamaParse if you write code and your documents are scans, photos, or pages full of complex tables and charts: its cloud service turns them into Markdown or JSON, and it is built for large volumes. Choose Agnos if you want a folder of text-based PDFs, Office files and saved web pages cleaned on your own Windows PC, with no code and no uploads, and a list of what it removed and the rule for each.

Gidion makes Agnos and wrote this page. Every claim about LlamaParse comes from LlamaIndex’s own website, documentation and GitHub, checked on 17 October 2026, and each source is linked under Sources. LlamaParse is the name of LlamaIndex’s whole document platform. This page compares Agnos with its parsing product, Parse, and mentions the rest of the platform where it matters. Statements about Agnos describe Agnos 2.3.2 and come from our documentation.

At a glance

AgnosLlamaParse
What it isA Windows app, made by Gidion, that cleans documents before they go into AI tools. It removes clutter with named rules and lists what it removed.A cloud document platform from LlamaIndex. Its Parse product uses generative AI models (except on its cheapest tier) to turn PDFs, scans, Office files and images into Markdown, text or JSON. The platform can also extract fields, classify, split and index documents.
Who it’s forPeople preparing files for Custom GPTs, NotebookLM, Claude Projects, Copilot, a RAG index or training data, without writing code.Developers and teams building document pipelines and AI agents. Its product page says it is “designed for developers”.
Where it runsOn your Windows PC. Documents are never uploaded; Agnos goes online only to activate, check or deactivate a licence.In LlamaIndex’s cloud, in North America or the EU. For enterprise customers, also as a dedicated single-tenant service, in their own AWS, Azure or Google Cloud account, or self-hosted on Kubernetes.
Needs codeNo. No Python to install, no terminal. The command line is coming to the Microsoft Store version in an update.Not for its web playground, which its docs suggest for non-technical users: it parses a file you upload and shows the result in the browser. For pipelines there is an API with SDKs for Python, TypeScript, Go, Java and .NET, a command line, and an n8n node you install from the n8n editor.
File typesText-based PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx), web pages, CSV, JSON, plain text, Markdown and source code. Old Office formats, images, audio, video and archives are refused.More than 130 formats, its docs say: PDF, Word, PowerPoint and Excel, including the old .doc, .ppt and .xls, RTF, EPUB, HTML, CSV, Apple’s Pages, Keynote and Numbers, images such as JPG, PNG, TIFF and HEIC, and audio such as MP3 and WAV.
Scanned PDFs (OCR)No OCR. A page with no text layer is skipped and named in the file’s warnings; a PDF with no text at all is refused.Yes. OCR runs by itself on any page with no text layer, on photos and on image files, and you can tell it which languages to expect.
TablesRead from Word, PowerPoint and web pages. In PDFs, only with the slower Layout reading mode. Excel sheets become tab-separated rows, without formulas or formatting. The RAG purpose keeps tables exactly as written.Recovered with their cell structure from PDFs, scans and images, including merged header cells and tables that continue across pages. Each comes back as Markdown or HTML and as rows and CSV, and optionally in an XLSX file.
What it outputsA RAG corpus in JSONL (an id, the source and the pages, slides, sheet or headings on each piece), cleaned .txt files, training data with an optional validation split, or all of it with a record of what was removed.Markdown, plain text, and per-page JSON with typed items such as headings, tables and images, with their positions on the page. Also tables as XLSX, page screenshots and images. Chunking and retrieval come with the platform’s Index product.
Shows what it removed, and whyYes. Review lists what it removed and the rule for each, up to 500 removals in a file, and the Everything export includes rejected.jsonl, with every removed piece and its reason.It returns each page’s detected header and footer in their own fields, apart from the page’s Markdown, and reports watermark text in a field of its own, so your code can keep or drop them. It also gives each page a confidence score.
Price14-day trial of everything in Pro, no card. Then $99 once for one person (2 devices), or $399 once for up to five people at one company (up to 10 devices), plus tax where it applies. No subscription.Credits, at $1.25 per 1,000. A page costs 1 to 45 credits by tier: 3 at Cost Effective, which its docs suggest for mostly-text documents with simple tables, and 10 at Agentic, which they suggest for scans, multi-column layouts and real tables. So 1,000 pages cost $3.75 at Cost Effective or $12.50 at Agentic. On Agentic and Agentic Plus, an optional Cost Optimizer sends a document’s simple pages to Cost Effective, and parsing the same file again with the same settings within 48 hours is free. Plans: Free with 10,000 credits a month, Starter at $50 a month with 40,000, Pro at $500 a month with 400,000, and Enterprise at a custom price.
Operating systems64-bit Windows 10 (version 2004 or later) and Windows 11, from the Microsoft Store.Nothing to install for the service itself: it runs in LlamaIndex’s cloud, and you reach it from a web browser or from your own code.
LicenceCommercial, under an end user licence agreement.Commercial. Its pricing page says LlamaParse “is not open source”. Its SDKs and command line are open source under the MIT License.

Where LlamaParse is the better choice

  • Your documents are scans or photos. Agnos has no OCR, so a scanned page gives it nothing to read. LlamaParse runs OCR on every page without a text layer and on image files, with a list of languages to expect. For a phone photo of a document, it can find the page, correct the perspective and even out the lighting before it reads it. Its docs list handwritten annotations and pages scanned at odd angles among the layouts Parse is built to handle.
  • Tables, charts and layout matter. LlamaParse keeps table cell structure, merged header cells included, can join a table that runs across pages, and can write every table to an XLSX file. Its docs recommend its Agentic tier for multi-column layouts and scanned pages. It can also turn charts into tables of data and return form fields with their checkbox states (in beta). For spreadsheets, it can find several tables in one sheet, recompute stale formulas and read hidden sheets. Agnos reads PDF tables only in its slower Layout mode, and turns Excel sheets into rows of plain text, without formulas and without hidden sheets.
  • You need to point to the exact source. Its bounding boxes go down to each word, line and table cell, so an app can highlight the passage an answer came from, and it can return the page number as printed on the page. Agnos records which pages, slides, sheet or headings each piece came from. LlamaParse can also return Word tracked changes and reviewer comments as structured data; Agnos does not read comments or tracked deletions.
  • Other kinds of file. Its list of supported formats runs past 130, including old .doc, .ppt and .xls files, RTF, EPUB, Apple’s formats, images and audio. Agnos asks you to save old Office files as .docx, .xlsx or .pptx first, and refuses images and audio.
  • You are building a pipeline. LlamaParse has an API with SDKs for Python, TypeScript, Go, Java and .NET, a command line, webhooks, batches of up to 10,000 files on its Pro and Enterprise plans, connectors that sync files from Google Drive, SharePoint, S3 and Confluence on plans that include folder syncing (on the Free plan you upload files yourself), and an n8n node. Agnos is a desktop app, and its command line is not in the Microsoft Store version yet.
  • You don’t use Windows. LlamaParse is a cloud service you use from a web browser or from your own code, so it does not depend on Windows. Agnos runs only on 64-bit Windows 10 and 11.
  • Scale, a team account and compliance papers. Its Free, Starter and Pro plans each take up to 100 users, and even a free account can be in its EU region. Enterprise adds SSO and up to 100 parse jobs at once. LlamaParse has completed a SOC 2 Type II audit, and offers a HIPAA pipeline with a BAA to Enterprise customers (Enterprise readiness). Agnos’s Company Pack covers up to five people, each in their own copy of Agnos: it is not a shared workspace or a server.
  • You need more than text. The same account can pull fields into a JSON schema you define (Extract), sort documents by type (Classify), split combined PDFs (Split) and run retrieval over your documents (Index).
  • Your volume is small. Its Free plan includes 10,000 credits a month: about 3,300 pages at Cost Effective, or 1,000 at Agentic, with no payment.

Want parsing from LlamaIndex on your own machine, and comfortable with a command line or code? LlamaIndex also publishes LiteParse, an open-source parser under the Apache 2.0 licence. Its README says it runs locally on Linux, macOS and Windows, with OCR through a bundled copy of Tesseract, and it points to LlamaParse for complex documents.

Where Agnos is the better choice

  • You don’t write code. Agnos is a Windows app from the Microsoft Store. You drop in files or a whole folder, check what was removed, and press Build and download. There is no Python to install: Agnos brings its own. LlamaParse’s web playground needs no code either, but its docs show batches, webhooks and folder-wide parsing done through its API, SDKs and command line.
  • Your documents must stay on your PC. Agnos reads, cleans and exports on your computer, and works with the network unplugged. Only the licence goes online, and no document text is sent. LlamaParse’s managed service works on LlamaIndex’s servers; its docs say files uploaded for parsing are cached there for 48 hours and then deleted, and that caching can be turned off. Files you store in its Files API with no expiry date stay until you delete them. Its docs also say it uses models from OpenAI, Anthropic and Google, under zero-data-retention and no-training agreements with those providers, and that customer data is not used for training by default. To keep documents in your own cloud account, LlamaParse offers bring-your-own-cloud and self-hosted deployments on Enterprise plans; there, its calls to AI models go from your own cluster to OpenAI, Anthropic or Google, or to their models on Azure, AWS Bedrock or Google Vertex AI, under your own keys.
  • You want to see what was removed. Review lists what Agnos removed and the rule for each piece, and Keep this anyway puts back a piece removed for its length, its quality or as a repeat. Lines removed while cleaning, such as page numbers and cookie notices, come back only if you switch their rule off and clean the file again. The Everything export adds rejected.jsonl, with every removed piece and its reason. If one file has more than 500 removals, Review counts them all but lists the first 500.
  • The clutter is in the text, not only at the edges of the page. LlamaParse already separates page headers and footers, can crop page edges, strip navigation and fixed elements from web pages, remove watermark text, skip hidden or diagonal text, and take a custom prompt that says what to skip. Agnos’s rules also remove cookie banners, “enable JavaScript” notices, placeholder text, page markers, forum clutter, and passages repeated within a file or across files. Review names the rule behind each removal, and the Everything export lists text dropped as a repeat of another file in duplicates.jsonl.
  • You have a folder of mixed files for an AI tool. One project takes text-based PDFs, Word, PowerPoint and Excel files, saved web pages, CSV, JSON and text together, subfolders included. Agnos exports a RAG corpus in JSONL, cleaned .txt files for Custom GPTs, NotebookLM, Claude Projects or Copilot, or training data with an optional validation split, made by file.
  • You’d rather pay once than per page. Agnos is $99 once for one person, or $399 once for up to five people at one company, after a 14-day trial with no card. There is no subscription and no export limit, and every 2.x update is included. LlamaParse charges credits per page, with monthly plans from $0 to $500, pay-as-you-go beyond a paid plan’s credits, and custom Enterprise pricing.

Using them together

LlamaParse returns Markdown, and Agnos reads .md files, so LlamaParse can go first for the files Agnos cannot read, such as scans, photos and old .doc files. Agnos then cleans that Markdown with the rest of your documents, lists what it removed and builds the export.

  1. Parse with LlamaParse. Ask for Markdown tables: set output_tables_as_markdown to true. By default LlamaParse can write a table as HTML, and Agnos’s Remove HTML tags setting would take out the tags and leave the cell text without its rows and columns. Review would not flag that: stripped tags show only in Preview’s Before and After panes. Markdown tables cannot show merged cells. If your pages carry a watermark, also set watermark_handling to remove: by default LlamaParse keeps its text at the end of each page. Save each result as a Markdown file, built from the per-page Markdown: LlamaParse returns each page’s detected header and footer in separate fields, so they stay out of the file.

  2. Add the Markdown files to Agnos. Drop the .md files, or their folder, on the Import step of a project with the RAG / Retrieval Corpus purpose, alongside your other documents. That purpose keeps headings, Markdown tables and lists as written.

  3. Review and export. Open Review, choose a file and press Removed, then build the export. By default, text found word for word in several files goes into a RAG corpus once, and each piece from a Markdown file carries the headings it sits under.

Two things to know:

  • For a text-based PDF, give Agnos the PDF itself. Agnos finds running headers and footers at the edges of each PDF page and records the pages each piece came from. A Markdown file has no pages, so that rule has nothing to look at; Agnos still removes a page marker such as “Page 3 of 10” on a line of its own.
  • The files you parse with LlamaParse go to its cloud at that step, and use its credits. Agnos’s part stays on your PC.

Each tool does its own job well. LlamaParse turns hard documents, scans included, into structured Markdown and JSON for code and AI models. Agnos cleans a folder of text-based documents on your own Windows PC and shows what it took out. If the second job is yours, you can try Agnos free for 14 days, with no card. What it does not do is listed in What Agnos doesn’t do.

Sources

Checked 17 October 2026. LlamaParse’s plans, prices and features change, so check the linked page before you decide.

LlamaParse

Agnos

LlamaParse 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.