Humata
ProductivitySearch, summarize and compare uploaded documents with source-linked answers. Humata adds team-oriented document access and page-based plans for ongoing knowledge work.
Questions across filesCited passages
MINERU / GIVE DOCUMENTS A READABLE STRUCTURE
MinerU
A document parsing engine that turns PDFs, scans and supported office files into structured content for AI workflows. Local use avoids a mandatory hosted subscription, with hardware requirements and additional license conditions to review.
MinerU official website captured October 3, 2026; the website and local software have separate service conditionsExplore the image ↓
What you can do with it
What this tool is for
MinerU prepares documents for tasks that need more than a flat copy of their text. It can identify layout and extract material into forms that are easier for an AI application to search, read and cite. This is useful when a PDF contains several columns, tables, formulas or figures whose relationship matters to the meaning of the page.
It is a parsing layer rather than a complete answer engine. A clean extraction helps downstream retrieval and analysis, but it does not establish whether the source itself is correct. The practical objective is to retain enough structure and location information that a later answer can be traced back to the relevant document instead of becoming an unsupported summary.
01
Prepare a research paper
02
Build a knowledge-base input
03
Recover a scanned report
04
Feed an agent a long document
A closer look at the product
Official product image. Click to inspect the details.
Product interface and examples
Key features, practical tasks
Capabilities in practice
- Layout-aware reading: Parse complex pages while retaining distinctions between text, tables and other elements. Review reading order in multi-column documents, because a sequence that looks plausible as plain text can still join unrelated paragraphs.
- Scanned document processing: Supported pipelines can use OCR to recover text from scans and images. Choose a clear source and check low-contrast characters, stamps and small print rather than assuming that successful extraction means every character was recognized correctly.
- Tables and formulas: MinerU targets structured tables and mathematical content. Compare row and column relationships, merged cells and equation symbols against the original page before using the output for calculations or research notes.
- Machine-readable exports: The current output contract describes multiple rendering targets, including Markdown and structured content. Different interfaces expose different subsets, so select an export supported by the exact CLI or API route you are using.
- Page and block references: The current document-reading workflow provides stable locators for pages and blocks. Retaining these references helps a downstream system return to the evidence instead of storing only text fragments with no connection to their source.
- Local quality choices: The current local implementation offers parsing tiers with different models and hardware needs. A lighter tier may be a sensible first pass, while a difficult page can justify a more capable configuration after you inspect the initial result.
- Agent integration: The project describes document discovery, caching, search and continued reading for agent workflows. Integrate these capabilities with a clear permission boundary so a parsing tool only reads the files intended for the task.
Get started, step by step
- 01
Select a representative document
- 02
Match the local setup
- 03
Inspect extraction quality
- 04
Connect the reviewed output
A practical first workflow
A practical first workflow
- Select a representative document. Choose a file that includes the structures you actually need, such as a table or a multi-column page. Use a permitted local copy and preserve it as the reference.
- Match the local setup. Follow the current installation guide and choose a parsing tier that fits available memory and acceleration. Download required models before expecting offline processing to work.
- Inspect extraction quality. Compare the output with the original at several difficult locations. Check reading order, table alignment and formula symbols, recording which classes of page require a different configuration.
- Connect the reviewed output. Export a supported format and retain source locations and version information. Only then feed the material into a search index, knowledge base or agent workflow.
Example prompt or task: Parse this public report into structured text, preserve page references, and flag tables or formulas that need visual review before the content enters a knowledge base.
This is a proposed processing workflow. The page does not claim a measured extraction accuracy or speed on your documents.
Find the right plan for you
Free local software
Explore official plans ↗Free software Your hardware; hosted services are separate
The current license is based on Apache 2.0 with additional commercial-threshold and attribution conditions.
What is free and what can cost extra
The local software provides a free route without a mandatory hosted subscription. You supply the computer, storage and any necessary acceleration; optional remote parsing and hosted services should be evaluated under their own terms rather than assumed to be unlimited and free.
The current license is based on Apache 2.0 with additional conditions. It requires attribution for online services provided to third parties and a separate commercial license if the specified consolidated thresholds are exceeded: more than 100 million monthly active users or more than US$20 million total monthly revenue. Review the actual license for your deployment.
A few things worth knowing
Frequently asked questions
Is MinerU a chatbot?
Its core role is document parsing; another application can use the extracted content for answers.
Can it run locally?
Yes, the project documents local pipelines and hardware choices.
Does local use require an API subscription?
Not for the local route, although hardware and optional external services can cost money.
Can it handle scanned pages?
Supported OCR pipelines can, but recognition quality still needs review.
Are all exports available everywhere?
No. The output contract says interfaces expose different subsets.
Is the license plain Apache 2.0?
No. The current license includes additional attribution and commercial-threshold conditions.
Should extracted numbers be trusted automatically?
No. Compare critical values and table relationships with the source page.
The details behind your decision
Explore platforms, inputs and outputs, licensing, and access requirements.
Four useful starting projects
Four useful starting projects
- Prepare a research paper: Extract a paper with columns and equations, then inspect one representative formula and table. Preserve page references so later notes can link back to the original argument.
- Build a knowledge-base input: Convert a small set of approved manuals into structured text before chunking them. Remove obsolete duplicates and keep the document version alongside each extracted record.
- Recover a scanned report: Use OCR on a readable scan and compare names, dates and numeric tables. Mark uncertain passages for manual review instead of quietly treating the output as authoritative data.
- Feed an agent a long document: Let an application retrieve relevant pages or blocks rather than inserting an entire archive into one prompt. Keep the source location attached to each passage used in the answer.
Platforms, inputs, outputs and setup
Platforms, inputs, outputs and setup
The current project supports PDFs, images and several office or web document formats, with different parsing tiers and export availability. Plain text does not need the same OCR path. Check the current format contract instead of applying one configuration to every file type.
Hardware requirements depend on the tier and inference backend. A CPU-compatible route exists, but that does not mean a large archive will process quickly on every laptop. Estimate capacity with a small representative sample and include model storage in the deployment plan.
Data, permissions and licensing
Data, permissions and licensing
Keep the local route local by checking the selected backend and whether remote parsing has been enabled. A locally installed application can still transmit documents when configured to use a hosted service.
Restrict access to source documents, extracted text and caches. Parsed output can make sensitive information easier to search, so protect the derived files as carefully as the originals and retain only what the downstream task needs.
Sources, documentation and help
Official sources and review date
Official sources and review date
Reviewed October 3, 2026. Product facts come from the official sources below. Suggested projects, prompts and review methods are FindGoodAI editorial guidance, not measured performance results.
Your experience can help someone choose
COMMUNITY
Sign in to rate and comment
Tried it? Share your experience
Loading experiences…
Manage my comments
Community and moderation
Saved tools are private. Approved comments are public; edits return to moderation. Comment counts include approved comments and replies. Each person has one rating. New ratings are automatically approved; withdrawn, unapproved or invalid ratings do not count. Share real experiences and avoid spam, private information or personal attacks.
For moderation appeals or data requests: support@findgoodai.com
NEXT STEP
Explore MinerU ↗Start with a difficult page, not an entire archive
Bring a real task and see how it fits the way you work.
Jump to a section
User experiences