Export datasets

Exports package a dataset into a ZIP archive you can download and use in other tools. The available formats depend on the dataset type.

Export formats

Dataset type	Format	What the archive contains
Images	COCO	An `images/` folder and an `annotations.json` file in COCO format
Documents	LLM Wiki	Original documents, their extracted text, and a wiki structure ready for an AI agent
Documents	Markdown ZIP	One Markdown file per document and its figures, without the original files

Exports include every ready asset in the source, whatever its review status.

COCO

Your labels become the COCO categories. Bounding boxes and polygons both export with a bounding box, a segmentation outline (the polygon, or the four corners of a box), and an area. Annota also records where each annotation came from and, for automatic annotations, its confidence.

LLM Wiki

A starting vault for the LLM Wiki pattern, where an AI agent builds an interlinked Markdown wiki from your sources. It also opens as an Obsidian vault.

Path	Content
`README.md`	What the archive contains
`AGENTS.md`	Instructions for AI agents working on the wiki (`CLAUDE.md` imports it)
`raw/sources/`	The original documents
`wiki/sources/`	One Markdown file per document, with your corrections and a link to its source
`wiki/index.md`	A catalog of the documents
`wiki/log.md`	A log of operations on the wiki

Office and text files are stored as converted Markdown, so their original in raw/sources/ is that Markdown.

Markdown ZIP

One Markdown file per document at the root of the archive, with figures in images/. Each file starts with its title, page count, and dates, and marks where each page begins.

When a document was only partly processed, both document formats list its missing page numbers at the top of the file.

Create an export

Open the dataset and select Exports.
Select Create export.
In Source, choose a version, for example Version 3, or Current dataset (live).
In Format, choose the format.
Select Start export.

Choose a dataset version for a reproducible export. The live dataset is read when the export runs, so edits made in the meantime can change the result.

The export is built in the background. Only one export can run at a time per dataset. If it takes at least a minute, you get an email when it finishes.

Status	Meaning
Queued	The export is waiting to start.
Running	Annota is building the archive.
Ready	The archive can be downloaded until the date on its row.
Failed	The export failed. Create a new export to try again.
Expired	The archive was removed after its expiration date.

Download or delete an export

When an export is Ready, select the download icon on its row. Download it before the Expires date shown on the row: after that date the archive is removed.

To remove an archive earlier, select the delete icon and confirm in Delete export?. This cannot be undone.

Costs and permissions

Exports use no credits. Saved archives count toward your organization’s storage until they expire or you delete them.
Anyone with view access to the dataset can create and download exports and delete the exports they created. People with edit access can delete any export. See Sharing and access.