Skip to content

Advanced Web Clipper Templates

Clip Templates are JSON files that control how the Web Clipper processes a web page: what metadata fields appear in the clip sheet, how page content is formatted into the document body, which pages the template activates on, and how images are handled.

Atlas uses the .atlasclip.json format, which is compatible with the Obsidian Web Clipper template format. The template language follows the engine Obsidian extracted from Web Clipper, so templates you copy from Obsidian's templates, variables, filters, and logic documentation render the same way in Atlas. Templates exported from Atlas can be imported into Obsidian, and Obsidian Web Clipper templates can be imported into Atlas.

Atlas adds a few extras on top, such as the atlas settings object, the round_to filter, and the on-device AI tags. If you have templates from an earlier version of Atlas, see Obsidian Compatibility and Older Templates.

Template Structure

Every .atlasclip.json template is a JSON file with the following top-level fields:

Field Type Description
schemaVersion string Use "0.1.0". Kept for Obsidian compatibility
name string Display name shown in the template picker
behavior string Kept for Obsidian compatibility. Atlas ignores it, because you choose the destination in the clip sheet
noteNameFormat string Template string for the document title, e.g. "{{title}}"
path string Kept for Obsidian compatibility. Atlas ignores it
properties array Metadata field definitions (see Metadata Properties)
triggers array URL or schema patterns that activate this template (see Triggers)
noteContentFormat string The body template (see Variables, Filters, Logic)
atlas object Atlas-specific extension fields (see Atlas Extension Fields)

Metadata Properties

The properties array defines the metadata fields that appear in the clip sheet's Details tab. Each item defines one field:

Field Type Description
name string Field label shown to the user
value string Default value or template expression for auto-population
type string Field type: text, multitext, number, checkbox, date, datetime, url

The value field can be a static string or a template expression. For example, "{{schema:@Recipe:author.name ?? author}}" to auto-populate an author field from Schema.org data with a fallback to the page's meta author.

Triggers

Triggers determine which template activates for a given page. Each trigger is a string in one of these forms:

Form Example Matches
URL prefix "https://github.com/" URLs that start with exactly this text
Regex "/^https:\\/\\/.*\\.wikipedia\\.org\\//" URLs matching the pattern. Write it between slashes, with no flags
Schema.org type "schema:@Recipe" Pages with Schema.org data of that type, including its subtypes (schema:@Article also matches a NewsArticle page)
Schema.org property "schema:@Recipe.recipeCuisine=Italian" Pages where that property has that value

URL prefixes are a plain "starts with" comparison against the full address. Atlas doesn't ignore www. or a trailing slash, so a trigger for https://github.com/ won't match https://www.github.com/. List each variant you want, and start every prefix with https://.

Which Template Wins

When more than one template could match a page, Atlas picks in this order:

  1. Your own templates are checked first, in alphabetical order by name. If none match, the bundled templates are checked. If none of those match, the Default template is used.
  2. Within each group, the template with the longest matching URL prefix wins. Ties go to the earlier template.
  3. If no prefix matches, the first template with a matching regex wins.
  4. If no regex matches, the first template with a matching Schema.org trigger wins.

Because your templates are checked first, a template of yours always beats a bundled one, even when the bundled template has a longer URL prefix.

Atlas Extension Fields

The atlas object inside a template contains Atlas-specific settings that extend the base format:

Field Type Description
synopsisFormat string Template for the Synopsis field
notesFormat string Template for the Notes field of a new document. When you append a clip to an existing document, its Notes are left alone
tagMappings array Automatic tag suggestions (see below)
imageHandling string Starting action for the clip's images (see below)
addWebPageChild boolean If true, "Save original page" defaults to on
icon string Kept with the template. Atlas doesn't use it yet
color string Kept with the template. Atlas doesn't use it yet

imageHandling sets where every image starts in the clip sheet. downloadAll saves them, linkAll links them, and skip removes them. prompt, or leaving the field out, keeps Atlas's per-image defaults. Images that look like page chrome, sidebar items, or tiny icons always start as Remove, and you can change any image before saving.

Tag Mappings

Each item in tagMappings has a source field that controls where tags come from:

Source Description
"metaKeywords" Uses the page's <meta name="keywords"> tag
"schemaType" Uses the Schema.org @Type of the page
"schemaCategory" Uses the Schema.org category field
"ogType" Uses the Open Graph og:type value
"contentKeywords" Extracts keywords from the article body
"fixed" Uses a hardcoded list; requires "fixedTags": ["tag1", "tag2"]

Variables

Variables are written as {{variableName}} and are replaced with values extracted from the page. Unknown variables resolve to an empty string and produce no error. Unknown filters are different: they produce an error, with a "did you mean" suggestion.

Lists and objects, such as {{highlights}} or a Schema.org array, render as compact JSON when you use them without a filter. Pass them through list, join, or table to get readable output.

Preset Variables

Variable Type Description
{{title}} string The page title
{{url}} string The full source URL
{{content}} string Extracted body as Markdown (see content cascade below)
{{contentHtml}} string Same cascade as {{content}} but raw HTML
{{description}} string Meta description tag content
{{author}} string Page author from meta tags or Schema.org
{{date}} string When the clip was taken, formatted as YYYY-MM-DDTHH:mm:ssZ in your local time zone, e.g. 2026-10-28T15:15:31-05:00
{{time}} string Same value as {{date}}
{{published}} string The page's publication date, when the page provides one
{{language}} string The page's language, when the page provides one
{{noteName}} string The page title with characters that aren't safe in file names removed
{{image}} string URL of the page's main or hero image
{{site}} string Site name from og:site_name or equivalent
{{domain}} string Domain only, e.g. github.com
{{favicon}} string Absolute URL of the page's favicon
{{selection}} string User's selected text on the page, as Markdown
{{selectionHtml}} string User's selected text on the page, as HTML
{{highlights}} array Array of highlight objects; each has text and timestamp properties
{{words}} number Word count of the article body
{{fullHtml}} string Complete raw HTML of the entire page
{{transcript}} string Transcript text for video pages (YouTube, Vimeo, Dailymotion), when one is available, broken into paragraphs. YouTube transcripts include a ### heading for each chapter listed in the video description. Empty otherwise
{{transcriptLanguage}} string Language code of the transcript
{{transcriptSource}} string How the transcript was made: manual (creator captions), asr (auto-generated), or translated
{{video}} object Video details, when the page is a supported video. Empty otherwise, so {% if video %} works as a check

Both {{date}} and {{time}} include the time of day. When you want only the date, format it: {{date | date:"YYYY-MM-DD"}}.

The {{video}} object has these fields, which you read with a dot:

Field Description
{{video.channelName}} Channel or uploader name
{{video.channelURL}} Link to the channel
{{video.videoID}} The site's identifier for the video
{{video.duration}} Length in seconds. Use {{video.duration | hms}} for a clock format such as 10:42
{{video.publishedDate}} Publication date as an ISO 8601 timestamp
{{video.viewCount}} Number of views

Not every field is present for every video. A missing field resolves to an empty string.

Content cascade: {{content}} follows a priority order rather than always returning the full article:

  1. If the user highlighted text before sharing — use only the highlights
  2. Otherwise, if the user selected text before sharing — use the selection
  3. Otherwise — use the full extracted article body

This means selecting a specific passage before tapping Share will clip only that passage.

Meta Tag Variables

Access any HTML <meta> tag using the meta: prefix:

{{meta:og:image}}       → content of <meta property="og:image">
{{meta:author}}         → content of <meta name="author">
{{meta:keywords}}       → content of <meta name="keywords">
{{meta:description}}    → same as {{description}}

Schema.org Variables

Access JSON-LD and Schema.org structured data embedded in the page using schema:@Type:path:

{{schema:@Recipe:author.name}}               → Author name from Recipe schema
{{schema:@Recipe:recipeIngredient}}          → Array of ingredient strings
{{schema:@Recipe:recipeInstructions[*].text}} → Extract .text from each instruction
{{schema:@Book:image ?? image}}              → Book cover image, fallback to meta image
{{schema:@Product:offers.price}}             → Product price

Path syntax:

  • . for nested property access: author.name → name inside the author object
  • [*] for array extraction: recipeInstructions[*].text → array of text values from each item

CSS Selector Variables

Extract content from the page HTML directly using CSS selectors:

{{selector:h1}}                  → Text content of the first <h1> element
{{selector:.article-author}}     → Text content of .article-author element
{{selectorHtml:article}}         → Inner HTML of the <article> element

Fallback Chains

Use ?? to try alternatives left-to-right, returning the first value that isn't empty, null, false, or 0:

{{schema:@Recipe:image ?? image}}
{{schema:@Book:author.name ?? schema:@Book:author ?? author}}

AI Variable

The prompt variable generates text with on-device AI. See AI-Powered Tags.


Filters

Filters transform a variable's value using the pipe operator |. They chain left-to-right:

{{title | upper | trim | truncate: 60}}

Atlas includes about 80 filters. Names are lowercase with underscores, such as date_modify and strip_tags. An unknown filter name produces an error and passes the value through unchanged, so a typo shows up in the problems list instead of silently doing nothing.

The tables below cover the most useful filters. The Obsidian filters documentation describes the full set, and everything there works in Atlas.

Text Filters

Filter Example Result
upper {{"hello" | upper}} HELLO
lower {{"HELLO" | lower}} hello
capitalize {{"hello world" | capitalize}} Hello world
title {{"hello world" | title}} Hello World
trim {{" hello " | trim}} hello
replace: "a":"b" {{"foo baz" | replace: "foo":"bar"}} bar baz
safe_name {{"File/Name?" | safe_name}} FileName
truncate: N {{"hello world" | truncate: 7}} hello …
truncatewords: N {{"one two three" | truncatewords: 2}} one two…
camel {{"hello world" | camel}} helloWorld
pascal {{"hello world" | pascal}} HelloWorld
snake {{"hello world" | snake}} hello_world
kebab {{"hello world" | kebab}} hello-world
uncamel {{"helloWorld" | uncamel}} hello world

replace takes one or more "search":"replacement" pairs. A search written as /pattern/flags is a regular expression.

safe_name removes characters that aren't safe in file names instead of swapping them for another character. Add "windows", "mac", or "linux" to apply only that system's rules. If nothing is left, the result is Untitled.

truncate counts the trailing … toward the limit, so truncate: 7 keeps six characters and the ellipsis.

Number and Math Filters

Filter Description
length Number of items in an array, or characters in a string
round Round to nearest integer
round: N Round to N decimal places
round_to: N Round to the nearest multiple of N. {{words | round_to: 1000}} turns 81234 into 81000
number_format Add thousands separators: 81000 becomes 81,000
calc: "expr" Apply arithmetic: number | calc: "*2" (supports +, -, *, /, and ^ for powers)
sum Add the numbers in an array
decode_uri URL-decode a string

round_to is an Atlas addition. Use it for approximate figures, such as a manuscript's word count to the nearest thousand.

Array Filters

Filter Description
first First item of the array
last Last item of the array
join: "," Join array items with a separator string
split: "," Split a string into an array by separator
slice: 1,3 Extract a sub-array, counting from 0 and stopping before the end position
nth: N Keep the Nth item (counting from 1). nth: 2n keeps every second item, and nth: n+3 keeps everything from the third item on
map: path Extract a property from each item: map: author.name. An arrow function also works: map: x => x.name
sort, reverse, unique, compact Sort, reverse, remove duplicates, or remove empty items
where: "key","value" Keep items whose property equals a value
merge: "a","b" Append values to an array
object: keys Turn an object into a list of its keys, values, or array of key-value pairs

Markdown Output Filters

Filter Output
blockquote > text
callout: ("type", "title") Obsidian-style callout block
footnote Markdown footnote references
fragment_link Link to the matching passage on the source page
image: "alt" ![alt](url)
link: "text" [text](url)
wikilink [[page]]
list Unordered Markdown list from an array. Use list: numbered, list: task, or list: numbered-task for other styles
table Markdown table from structured data. table_pretty aligns the columns
bold, italic, strike, highlight, code Inline formatting
h1 to h6 Heading of that level
yaml Format a value as YAML
template Fill ${} placeholders from each item: "- ${title}"

HTML Processing Filters

Filter Description
strip_tags Remove all tags and keep the text. strip_tags: "b" keeps the named tags
remove_tags: "span" Remove the named tags but keep their content
remove_html: "script,.ad" Remove the elements that match CSS selectors, along with their content
replace_tags: "b":"strong" Rename a tag, keeping content and attributes
strip_attr: "href,src" Remove every attribute except the ones you name
remove_attr: "onclick" Remove the named attributes and keep the rest
strip_md Remove Markdown syntax, leaving plain text
markdown Convert HTML to Markdown

Date Filters

Filter Description
date: "format" Format a date with tokens such as YYYY, MMMM, D, HH, mm, and ss. Add the format the date is written in as a second value when Atlas can't read it automatically: date: ("YYYY-MM-DD", "DD/MM/YYYY")
date_modify: "+1 day" Add or subtract time (seconds, minutes, hours, days, weeks, months, years). The result is a YYYY-MM-DD date
duration: "format" Format a number of seconds, or an ISO 8601 duration such as PT2H30M, with tokens such as HH:mm:ss
hms Format seconds as a clock time: 642 becomes 10:42
first_paragraph Keep only the text before the first blank line. Useful for turning a long description into a synopsis

Dates use your device's local time zone.

AI Filter

The summarize filter condenses text with on-device AI. See AI-Powered Tags.


Logic

Conditionals

Use {% if %} / {% elseif %} / {% else %} / {% endif %} to include content conditionally:

{% if author %}
By {{author}}
{% elseif site %}
From {{site}}
{% else %}
Author unknown
{% endif %}

Truthiness: A value is falsy if it is null, false, 0, or "". An empty list [] also counts as empty. An empty object {} is truthy. Everything else is truthy.

Comparison operators:

{% if author == "Jane Doe" %}    → equality
{% if count != 0 %}              → inequality
{% if price > 10 %}              → greater than
{% if price < 100 %}             → less than
{% if score >= 90 %}             → greater than or equal
{% if length <= 5 %}             → less than or equal
{% if tags contains "recipe" %}  → text or list contains a value (ignores case)

Combine conditions with and, or, and not. The symbols &&, ||, and ! work the same way:

{% if author and not selection %}
{% if site == "GitHub" or site == "GitLab" %}

contains checks text or lists. The word in is only for for loops, so {% if x in list %} is an error.

Add notes to a template that won't appear in the output with {# comment #}. The whitespace-trimming forms {%- and -%} aren't supported.

Loops

Use {% for item in collection %} / {% endfor %} to iterate over an array:

{% for item in schema:@Recipe:recipeIngredient %}
- {{item}}
{% endfor %}

A loop object is available inside each iteration:

Property Description
loop.index Current iteration number, 1-based
loop.index0 Current iteration number, 0-based
loop.first true if this is the first iteration
loop.last true if this is the last iteration
loop.length Total number of items

Example with numbered steps:

{% for step in schema:@Recipe:recipeInstructions[*].text %}
{{loop.index}}. {{step}}
{% endfor %}

Each pass is trimmed, empty passes are skipped, and the results are joined with line breaks. You can only loop over lists. Looping over an object is an error, so convert it first with object: keys.

Variable Assignment

Use {% set name = value %} to create a local variable:

{% set slug = title | kebab %}
File: {{slug}}.md

{% set displayDate = date | date: "MMMM D, YYYY" %}
Saved on {{displayDate}}

Set variables can use filters and are scoped to the template render.


AI-Powered Tags

Two tags call on Apple Intelligence to generate text on-device as you clip. Both require a device that supports Apple Intelligence — an iPhone, iPad, or Mac running iOS, iPadOS, or macOS 26 or later with Apple Intelligence turned on. All processing stays on the device; the page content never leaves it.

summarize

A filter that condenses any text value into a short summary:

{{content | summarize}}        → a roughly 280-character summary of the body
{{content | summarize: 150}}   → a summary capped near 150 characters

Pass an optional length, in characters, as the argument. summarize works on any text value — {{content}}, a selector result, a schema value — not just the body. On devices without Apple Intelligence it falls back to trimming the text with an ellipsis, so the template still renders. That makes it a natural fit for synopsisFormat:

"synopsisFormat": "{{content | summarize}}"

prompt

A variable that sends a written instruction to the on-device model and inserts its reply:

{{prompt:"Write a one-sentence, friendly reminder to read this article later."}}

Use single or double quotes around the instruction. The instruction is sent on its own — page variables aren't fed into it automatically — so prompt is best for instruction-only text; to summarize the clipped content itself, use summarize. On devices without Apple Intelligence, prompt resolves to an empty string.

Note

Because these tags generate text while you clip, Atlas resolves them once up front — you may see a brief loading state — and reuses the result. Long inputs are trimmed before processing, and responses are inserted as plain text.

Obsidian Compatibility and Older Templates

Atlas follows Obsidian Web Clipper's template rules, so a template that works in Obsidian works here. You can import Obsidian templates, including several files at once and Obsidian's whole-settings exports. When you export a template, choose Atlas Template to keep the atlas settings, or Obsidian Web Clipper for a file Obsidian can read. See Web Clipper for the editor itself.

Atlas checks a template when you save it and lists any problems. A red badge on a template in the list means an error, and an orange badge means a warning.

If you wrote templates for an earlier version of Atlas, a few things changed:

Before Now Handled for you
{% elif %} or {% else if %} {% elseif %} elif only
strip_markdown strip_md Yes
x in list in a condition list contains x No
map:"{{ item }}" map: path or map: x => x.name No
template_string template with ${} placeholders No
replace: "a","b" replace: "a":"b" No
date: "format", "timezone" date: "format" (dates use your local time zone) No
/pattern/i triggers A regex without flags, or a URL prefix No
schema:@Type:key=value schema:@Type.key=value No
Domain triggers such as nytimes.com A full prefix such as https://nytimes.com/ No

Atlas renames the first two for you when it loads your saved templates, and keeps a backup of the original next to the template file with a .pre-rosetta2 ending. The remaining changes need a manual edit. Atlas flags them as problems when you save the template.

Full Example

A complete Recipe template:

{
  "schemaVersion": "0.1.0",
  "name": "Recipe",
  "behavior": "create",
  "noteNameFormat": "{{schema:@Recipe:name ?? title}}",
  "path": "Recipes/",
  "properties": [
    { "name": "author",    "value": "{{schema:@Recipe:author.name ?? author}}", "type": "text" },
    { "name": "prep_time", "value": "{{schema:@Recipe:prepTime}}",              "type": "text" },
    { "name": "yield",     "value": "{{schema:@Recipe:recipeYield}}",           "type": "text" }
  ],
  "triggers": ["schema:@Recipe"],
  "noteContentFormat": "# {{title}}\n\n{{description}}\n\n## Ingredients\n\n{{schema:@Recipe:recipeIngredient | list}}\n\n## Instructions\n\n{{schema:@Recipe:recipeInstructions[*].text | list:numbered}}",
  "atlas": {
    "synopsisFormat": "{{description}}",
    "tagMappings": [
      { "source": "schemaType" },
      { "source": "fixed", "fixedTags": ["recipe", "cooking"] }
    ],
    "imageHandling": "prompt"
  }
}

See Also