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
DOCLING / ONE STRUCTURE FOR MANY DOCUMENTS
Docling
A local document-processing toolkit that converts varied files into a common structured representation for search and generative AI. It combines PDF layout understanding, OCR and developer-friendly exports under an MIT software license.
Official Docling processing diagram from the project repository, illustrating the document conversion pipelineExplore the image ↓
What you can do with it
What this tool is for
Docling is a toolkit for turning different document formats into a consistent representation that other software can use. It addresses the preparation stage of an AI system: understanding page layout, recovering text and tables, and exporting content without forcing every downstream application to handle each source format from scratch.
This is useful for mixed collections containing PDFs, office documents, web content and scans. Instead of equating conversion with a plain-text dump, Docling provides a structured document model. That structure helps developers keep relationships between elements and makes it easier to inspect what entered a search index or a retrieval-augmented generation pipeline.
01
Normalize a document collection
02
Prepare retrieval chunks
03
Inspect scanned archives
04
Extract a table for analysis
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
- Multiple input formats: The current project supports a broad range of document formats, including PDF, office files, HTML and images. Check the supported-format guide for the exact version and pipeline; a supported extension does not guarantee identical extraction quality for every internal layout.
- PDF layout analysis: Docling understands page organization and reading order rather than relying only on embedded text. Inspect sidebars, footnotes and multi-column pages, where otherwise readable text can be placed in the wrong sequence.
- Table structure: Table extraction aims to preserve relationships between cells. Before importing numbers into a dataset, verify headings, merged cells and units against the original, especially when a table continues onto another page.
- OCR options: Scanned PDFs and images can be processed through supported OCR engines. Language models and engine configuration matter, so a Chinese scan should be checked with an appropriate setup rather than judged using an unrelated default.
- Unified document representation: DoclingDocument provides a common structure for extracted content. Keeping that structured form can be more useful than immediately flattening everything into Markdown, particularly when a later task needs element types and source relationships.
- Export and integration: The project supports outputs such as Markdown, HTML and JSON and integrations with AI frameworks. Choose the format that preserves the information your next step needs, and keep an original structured result for debugging.
- Local and service workflows: You can use a command line or Python integration, and the project documents API-server and MCP options. These are deployment choices, not a requirement to publish your documents to a public endpoint.
Get started, step by step
- 01
Choose a representative sample
- 02
Install a suitable local setup
- 03
Convert and compare
- 04
Integrate gradually
A practical first workflow
A practical first workflow
- Choose a representative sample. Include a normal page and a difficult page from the intended workload. Keep the originals and decide which output details, such as tables or headings, must survive conversion.
- Install a suitable local setup. Follow the current installation guide for your operating system and Python environment. Prepare model downloads and optional OCR dependencies before trying to run without network access.
- Convert and compare. Run the chosen pipeline, then inspect the structured output and a human-readable export. Check reading order, table cells and text around figures rather than looking only at whether the command completed.
- Integrate gradually. Connect reviewed results to the next application and log file-level failures. Keep version and configuration information so a later parser upgrade can be compared against the previous output.
Example prompt or task: Convert a mixed public document sample into structured content and Markdown. Preserve headings and tables, and record any pages whose reading order needs manual review.
This is an editorial validation task, not a claim that the toolkit converts every source with identical accuracy.
Find the right plan for you
Free software
Explore official plans ↗Free software Local compute and optional services are separate
The software license does not automatically cover every downloaded model, remote service or source document.
What is free and what can cost extra
Docling’s software is released under the MIT license and can be run locally without a mandatory service subscription. You still provide compute, model storage and maintenance, and a larger deployment can require acceleration or additional infrastructure.
Optional remote services, hosted integrations and downloaded models have their own conditions. The MIT software license should not be treated as a universal license for every model weight or source document used with the toolkit. Review these components separately when packaging or distributing a solution.
A few things worth knowing
Frequently asked questions
Is Docling free software?
Yes. The project’s software license is MIT.
Is it a finished chatbot?
No. It prepares document content that another application can search or use for answers.
Can it run locally?
Yes, including workflows suitable for offline environments after resources are prepared.
Does it handle scans?
It offers OCR support, with engine and language configuration to consider.
Is Markdown the only output?
No. The project includes a structured document model and several export formats.
Are all models covered by MIT?
Do not assume that; model and service licenses are separate from the toolkit license.
Should I validate extracted tables?
Yes. Check cell structure, units and critical values before analysis.
The details behind your decision
Explore platforms, inputs and outputs, licensing, and access requirements.
Four useful starting projects
Four useful starting projects
- Normalize a document collection: Convert a small folder containing several source formats into a common structure. Compare each format separately so a successful PDF conversion does not hide a weak spreadsheet or slide workflow.
- Prepare retrieval chunks: Use headings and content boundaries when dividing documents for search. Retain source metadata so a retrieved passage can be shown alongside the document that supports it.
- Inspect scanned archives: Apply OCR to a sample of scans and classify recurring errors. Use the findings to decide which pages need better scans or manual correction before processing the full collection.
- Extract a table for analysis: Recover a table, then check its numeric values and structure against the source. Treat the result as a draft dataset until units, totals and missing cells have been reviewed.
Platforms, inputs, outputs and setup
Platforms, inputs, outputs and setup
The installation guide supports macOS, Linux and Windows configurations, with dependencies and acceleration varying by platform. Choose a supported environment and avoid assuming that a demo command implies every optional pipeline is already installed.
Offline operation is possible with the necessary local resources in place. A first run may otherwise need model downloads. Record the selected OCR engine, pipeline and model versions if the output will feed a repeatable production or research process.
Data, permissions and licensing
Data, permissions and licensing
Local processing can help keep document contents under your control, but inspect whether any optional enrichment or remote model service is enabled. Keep network behavior consistent with the data policy for the collection.
Protect intermediate files and indexes as well as the original documents. Conversion can expose hidden or forgotten text to a search system, so apply document-level access rules when the result is consumed by a shared application.
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 Docling ↗Make document ingestion easier to inspect
Bring a real task and see how it fits the way you work.
Jump to a section
User experiences