Spacy NER annotation tool

A lightweight, browser-based tool for hand-labeling text spans and exporting them as spaCy NER training data — no backend, no build step, no install. Open index.html and start annotating.

Features

• Custom entity labels — define any number of labels (e.g. PERSON, ORG, DATE), each auto-assigned a random color chip.
• Keyboard shortcuts for labels — the first 26 labels you create are auto-assigned a letter, a through z, in order (shown as a small keycap on the chip). Select text in the sentence box and press the letter to tag it — no need to reach for the mouse. Labels beyond the 26th simply have no shortcut. Removing a label re-numbers the rest so the sequence stays gap-free.
• Click-to-tag annotation — select a span of text in the current sentence and click a label chip (or press its shortcut key) to tag it. Double-click a tagged span, or use the × next to its entry, to undo a tag.
• Raw data queue — paste text (one sentence per line) directly into the queue, or upload a .txt file. Sentences are annotated one at a time, in order, with a live progress bar and "sentence X of Y" indicator.
• Completed annotations panel — finished sentences accumulate here in spaCy's training tuple format. Copy them to the clipboard or download them as a .txt file at any point — you don't have to finish the whole batch first.
• Local, persistent, and safe to walk away from — everything (labels, queue, in-progress and completed annotations) is saved to the browser's localStorage as you work. You can close the tab or browser at any time and resume later on the same device/browser with nothing lost. An in-app note reminds users of this and can be dismissed permanently.
• Light/dark theme — follows your system preference by default, with a manual toggle that's remembered across visits.

Usage

1. Open index.html in a browser (works as a static file, or served via a local server such as XAMPP/Apache).
2. Under Entity labels, add the labels you want to annotate with (e.g. PERSON, LOCATION).
3. Under Raw data queue, paste your text (one sentence per line) or click Upload .txt to load a file. The sample count and progress bar are calculated from this queue.
4. The first sentence appears under Annotate sentence. Select a span of text, then either click a label chip or press its shortcut key to tag it as that entity. Repeat for every entity in the sentence — tagged spans and their (start, end, "LABEL") positions appear under Annotated entities.
5. Click Mark as completed to commit the sentence to the Completed annotations panel and advance to the next sentence in the queue.
6. Use Copy or Download .txt at any time to export what's been completed so far.

Output format

Each completed sentence is stored as a Python tuple in spaCy's standard NER training format:

("Barack Obama was born in Hawaii", {"entities": [(0, 12, "PERSON"), (25, 31, "LOCATION")]}),

The full downloaded/copied file is a sequence of these tuples, ready to drop into a spaCy training script (e.g. as the contents of a TRAIN_DATA list).

Data & privacy

All data lives in the browser's localStorage for this page's origin — nothing is uploaded or sent anywhere. That means:

• Progress persists across tab/browser closures on the same device and browser.
• Clearing browser data (or using a different browser/device) will lose anything not yet copied/downloaded.
• Use Copy or Download .txt whenever you want a durable backup of your completed work.

Project structure

index.html          Single-page application markup
css/main.css         Design tokens, layout, and component styles (light/dark theme)
js/main.js           Core annotation logic (labels, tagging, queue, import/export)
js/theme.js          Light/dark theme toggle
images/agate_logo.png  Brand mark used in the header

Tech stack

Plain HTML, CSS, and JavaScript (jQuery for DOM handling). No build tooling, package manager, or server-side component required — it's a static site that can be hosted anywhere or opened directly from disk.