Is flatmark a hosted Docling?

For the queue, yes. flatmark runs Docling there with caps of 25 MB, 200 pages and about 2 minutes of conversion per file. The direct call runs MarkItDown instead.

Should I self-host Docling instead?

If you can run it, that is a sound choice. Docling is MIT-licensed (source: github.com/docling-project/docling, as of Sep 2026). The trade is running its layout and OCR models on your own hardware.

Is the output identical to running Docling myself?

It is the same engine with flatmark's caps and settings: OCR on, table mode accurate. flatmark fails a file over 25 MB or 200 pages, or a conversion over about 2 minutes. Your own install might finish such a file.

flatmark vs Docling

Docling is the open-source converter behind flatmark's queue. Run it yourself, or send files to flatmark, which runs the same engine with its own caps.

Last updated Sep 24, 2026

Docling	Open-source (MIT) document converter started by IBM Research Zurich. Hosted by the LF AI & Data Foundation (as of Sep 2026)
Docling features	Page layout, reading order, table structure, formulas, OCR for scanned documents (as of Sep 2026)
Relationship	flatmark's queue (POST /v1/convert/jobs) runs Docling with OCR on and table mode accurate
flatmark caps	25 MB, 200 pages and about 2 minutes of conversion per file

Docling is an open-source Python toolkit for document conversion under the MIT license. IBM Research Zurich started it, and the LF AI & Data Foundation now hosts it (GitHub, as of Sep 2026). It reads PDF, Word, PowerPoint, Excel, HTML and more, with OCR for scans (same source).

flatmark's queue runs Docling. You send a file to POST /v1/convert/jobs and get Markdown plus the DoclingDocument JSON file. flatmark adds the API key, the queue, signed webhooks and an MCP server.

What differs from your own install

flatmark runs the same engine with its own caps: 25 MB, 200 pages and about 2 minutes of conversion per file. OCR is on and table mode is accurate. A partial conversion counts as failed, and the 10 credits are refunded.

When Docling fits

You can run Python and ML models on your own hardware. You want control over versions, models and settings, or files must stay on your machines. Docling costs nothing to use under MIT; the cost is the hardware and the operation.

When flatmark fits

You want Docling output without installing or operating it. One call queues the file. You poll GET /v1/jobs/{id} at no cost or receive a signed webhook. A queued job costs 10 credits and needs an API key.

FAQ

Is flatmark a hosted Docling?: For the queue, yes. flatmark runs Docling there with caps of 25 MB, 200 pages and about 2 minutes of conversion per file. The direct call runs MarkItDown instead.
Should I self-host Docling instead?: If you can run it, that is a sound choice. Docling is MIT-licensed (source: github.com/docling-project/docling, as of Sep 2026). The trade is running its layout and OCR models on your own hardware.
Is the output identical to running Docling myself?: It is the same engine with flatmark's caps and settings: OCR on, table mode accurate. flatmark fails a file over 25 MB or 200 pages, or a conversion over about 2 minutes. Your own install might finish such a file.