This site is free, but Google API usage is billed to your own account. Check quotas and pricing in Google AI Studio.

It is stored only in your browser's localStorage and never sent to our servers. You can verify in the network tab that requests only go to googleapis.com.

How long does it take?

Requests are spaced 15 seconds apart to stay inside free-tier limits, so a 100-page document at 5 pages per request takes over five minutes. You can set a page range; several images can go in one run, documents one at a time. Markdown and text files are accepted too, and an optional second pass evens out heading levels.

Multi-level headers are flattened with '>' and merged cells repeat the parent value. Three or more merge levels are simplified.

How are HWP and Word read?

HWP can be sent as extracted text (fast) or as page images (accurate). Word, Markdown and text files are always sent as extracted text.

AI Document Convert

Use your own Gemini API key to turn PDF, HWP, Word and images into high-quality Markdown, also saved as DOCX, HWPX or HWP. Multi-level merged tables, empty-cell padding and page markers are handled by default, and you can edit the prompt. Files and key go straight from your browser to Google.

DocMoa AI accepts PDF, image, HWP/HWPX, DOCX, Markdown, and text inputs. Your Gemini API key and either the files you select or content extracted or rendered from them are sent directly from your browser to Google’s generativelanguage.googleapis.com. DocMoa’s servers never receive or store this data. Its processing is governed by the Google Privacy Policy and the Gemini API Terms. Your API key is stored only in your browser’s localStorage and can be deleted at any time by selecting “Forget key.”

Gemini API key

Your key is stored only in this browser (localStorage) and is never sent to our servers. Requests go directly from your browser to Google.

If your key has HTTP referrer restrictions, add this site's domain to the allow list.

Get an API key (Google AI Studio)

When plain extraction is not enough

Plain text extraction reads the characters a PDF already stores, which is why it returns a jumble from a two-column paper, nothing at all from a scan, and a row of numbers with no shape from a table. A model that can look at the page works the other way round: it sees the layout and rebuilds it.

This runs on your own Gemini API key, kept in your browser's storage. Choose a model, how many pages go in each request, and a prompt: one preset concentrates on tables, another handles general documents, and you can write and save your own. Vision mode sends the pages themselves, PDF pages as they are and HWP pages as rendered images, which is what scans and complex layouts need; text mode sends extracted text instead, which is cheaper and useless for a scan. Chunks that fail are marked in the output rather than quietly dropped, so you can see what needs a second pass.

This is the one tool on the site that sends your document away. Pages go to Google's API under your own key and their terms, and usage is billed to you, so it is not the place for anything confidential. Every other conversion here stays inside your browser.

How to use

1
Enter your Gemini API key (stored only in your browser).
2
Pick files, a preset (table / general / custom) and output formats.
3
Click Start; chunks are converted and saved in the chosen formats.

Frequently asked questions

Does it cost money?: This site is free, but Google API usage is billed to your own account. Check quotas and pricing in Google AI Studio.
Is my key safe?: It is stored only in your browser's localStorage and never sent to our servers. You can verify in the network tab that requests only go to googleapis.com.
How long does it take?: Requests are spaced 15 seconds apart to stay inside free-tier limits, so a 100-page document at 5 pages per request takes over five minutes. You can set a page range; several images can go in one run, documents one at a time. Markdown and text files are accepted too, and an optional second pass evens out heading levels.
Complex tables?: Multi-level headers are flattened with '>' and merged cells repeat the parent value. Three or more merge levels are simplified.
How are HWP and Word read?: HWP can be sent as extracted text (fast) or as page images (accurate). Word, Markdown and text files are always sent as extracted text.