Skip to main content
Join the official Python Developers Survey 2018 and win valuable prizes: Start the survey!

Provides a way to store, relate to, extract metadata and embed content from external URLs via a modular backend system

Project description

TravisCI status Coverage status PyPI Version License

Armstrong.Apps.Embeds provides a data model and modular backend system to extract embeddable content and metadata from external URLs. Representing embeddable content in a database brings the typical benefits of relational data. Programmable backends allow more flexible use of the content beyond the one-trick pony of the standard “<iframe>” copy-paste embed code.

Integrating AppsEmbeds into your site will require some work. Mostly because this package doesn’t make assumptions for how you’ll be using these external URLs. Maybe you just want to track references to embedded content, maybe it’s the caching that’s interesting or programmatically accessing content provider metadata is what you are after. There’s nothing extra to get in your way, but you’ll have to customize code and/or templates.

This is a stand alone component; it does not require any other pieces of the Armstrong family. However, it does play nicely with ArmLayout. If you use ArmLayout already, AppsEmbeds is ready to go. Much of the power of AppsEmbeds comes from templating so if you use that feature and aren’t already using ArmLayout, it’s worth considering.

The second optional package is lxml, which is necessary if you use the resize_iframe template filter. Otherwise, this package has three fixed requirements to provide model fields and support the backend APIs (currently just Embedly). See package.json for these fixed requirements.


  • Single embed, multiple relationships! Always know where you are using external content and the number of references you have.
  • Metadata! Access the metadata of the external content. Much more useful than the vanilla iframe embed code.
  • Modular backends! Get metadata and embed codes with a standard interface to the various content provider APIs.
  • Customizable templates! Presentation is separate from data. Make that external content look appropriate for your site.
  • Single embed, multiple uses! Create templates for each use case. A video can be used as a title, thumbnail, captioned credit, in a gallery, etc.
  • Uniform visual appearance! Each content type shares templates. Video content can be presented in a standard format regardless of the source.
  • Automatic backend assignment! Program an Embed in one step. The URL is all we need.
  • Admin preview! Examine the response data before saving. Look weird? Switch backends and preview again.
  • Response caching! Worry less about third-party API failure. The response is cached indefinitely.
  • Check for new data! See if new data is available without committing to it, right from the Admin.
  • Bird’s eye overview! Aggregate information on how and what types of external content you generally use.

Installation & Configuration

Supports Django 1.3, 1.4, 1.5, 1.6, 1.7 on Python 2.6 and 2.7. (Though if you are using Django 1.3, make sure to use django-model-utils<1.4.)

  1. pip install armstrong.apps.embeds
  2. [optional] pip install lxml if you plan on using the resize_iframe template filter
  3. [optional] pip install armstrong.core.arm_layout if you want to use ArmLayout to render templates
  4. Add armstrong.apps.embeds to your INSTALLED_APPS
  5. Install the database schema
    • Django 1.7+ use migrate
    • previous Djangos use either syncdb or migrate if you are using South (in which case use South 1.0+)
  6. Load the provided Backends into your database. (This is not provided as initial fixture data so that you may edit them without worrying that syncdb will restore the initial versions.) loaddata embed_backends.json
  1. If you are using the Embedly backend, add your API key to
EMBEDLY_KEY = 'your key'

Logging: This component emits logging statements using the armstrong.apps.embeds logger.


A quick overview of the four models–

EmbedType and Provider do essentially nothing besides normalize the database and provide a quick way to perform aggregation queries. EmbedType is based on the four oEmbed types, though in practice you’ll likely have a fifth error type (and that’s okay).

Backend is the front-end model for the actual code that connects to third-party APIs to retrieve response data for the external content URLs. It’s easiest to initially load them from the fixture data file but feel free to customize them as you will. Just don’t change the slug, which is how the model maps to its code back-end. regex and priority are designed to change. That’s how you’ll customize the auto-assignment behavior. The Embedly backend will handle YouTube sure, but say you’ve written a more targeted YouTube-specific backend–add it to the database with a selective regex and a higher priority.

Embed is the cornerstone. Creating a new embed object only requires a url. The backend will auto-assign by regular expression matching the URL and selecting from the matching backends by highest priority. Auto-assignment is just a nice feature for faster Embed creation. You can also manually assign a backend on object creation or later. Every other field on the model is backend-provided metadata. Consider them read-only. So how do you get a response? How do you get actual information?

The Response object–

embed_obj.update_response() will retrieve a response from the backend and assign it if the response is valid and different, then return True. If the response is invalid or the same, no assignment is made and False is returned. If you want the response object itself, use embed_obj.get_response().

embed_obj.response is the way to access the response data. This will be a subclass of the BaseResponse object with a standard set of attributes. is_valid() will be False in cases where the API had a problem, didn’t return data, 404’d, etc. is_fresh() will be True when the response is fresh off the wire. It’s used to differentiate from database cached response data and you can probably ignore it. type and provider are EmbedType and Provider model objects. _data holds the actual raw response in JSON. The goal is to never directly access this. Instead, the BaseResponse class is subclassed by each backend/API and tailored to parse the raw data into standardized attributes. This way who cares if it’s YouTube or Vimeo, access the object the same and share the templates. These attributes return an empty string when nothing is available and are therefore template safe. The current data attributes are:

  • title
  • author_name
  • author_url
  • image_url
  • image_height
  • image_width
  • render

render is perhaps the most important; it is the full expression of the embedded content that the content provider offers. For Twitter, this is the blockquote with JavaScript widget that dynamically loads the tweet into an iframe. For YouTube and Vimeo, this is the video player. Whatever way the service designs its content to be embedded, this is it.

image_xxx means different things depending on the content. For a video, this will be the still image that shows before the video is played. For SlideShare, it’s the first slide in the presentation. For Flickr, it’s the thumbnail. It’s worth noting that we have no idea what the image size will be and so if you use this in a template, consider fixing the image tag’s dimensions with attributes or CSS.


Embedly is a sort of meta-embed service. They know how to handle over 250 content providers to deliver a standardized set of metadata. Specifically this backend uses their “Embed” service via their embedly-python library. It offers a huge benefit but does require an account. Fortunately there is a quite reasonable free tier. Configuration required to use this is mentioned under the Installation section.

Twitter is a simple wrapper for a tag that loads the tweet via Twitter’s JavaScript widget. It does not perform any API or network calls and therefore does not provide any metadata about the URL. The only thing it can do is embed the Tweet as if you’d copy-pasted the embed code.

Default just regurgitates the provided URL. It’s the catch-all that does nothing useful.


Assuming you want to display the embed content on your site, this is where you’ll spend the most developer time. It’s not just about what a photo looks like versus a video. Now that you have access to more than just the “embed code”–now that you have metadata–you can use the same embed multiple ways. For example, a photo can be used as a preview thumbnail with a small image, a larger image with a title for lead art, a thumbnail in a story that expands into a modal full-size version with attribution. Whatever you want. Since Response objects have a standard interface, it doesn’t even matter where that photo came from. Instagram and TwitPic behave the same.

Note: This concept of provider apathy hinges on the EmbedType. We can only treat like types the same or fall back to something generic for all embeds. If the provider or the backend reports a Flickr URL as a “link” type, even though we know in our hearts it’s a “photo”, it won’t use the photo-specific templates.

Now for some examples. Since ArmLayout was designed for this purpose, we’ll use it. It provides a render_model template tag that takes an object and a template name then looks in a hierarchy from most-specific to least for that template. ArmLayout uses get_layout_template_name() for the lookup and AppsEmbeds has extended it to also look for type-specific templates.

render_model embed_obj 'full' for a photo type will look in this order:

  • layout/embeds/embedtype/photo/full.html
  • layout/embeds/embed/full.html

So to display an Embed object as “preview”, just make the following files. Each content type can customize what “preview” means. (Maybe a small thumbnail or truncated intro text.)

  • layout/embeds/embedtype/photo/preview.html
  • layout/embeds/embedtype/video/preview.html
  • layout/embeds/embedtype/link/preview.html
  • layout/embeds/embedtype/rich/preview.html
  • layout/embeds/embed/preview.html

“Lead art” could be another way of displaying an embed. (Perhaps a larger image along with title and author attribution.)

  • layout/embeds/embedtype/photo/lead_art.html
  • layout/embeds/embedtype/video/lead_art.html
  • layout/embeds/embedtype/link/lead_art.html
  • layout/embeds/embedtype/rich/lead_art.html
  • layout/embeds/embed/lead_art.html

Leave out a type-specific template file and ArmLayout will use the more general file next in the hierarchy.

There’s also a default.html template used as a fallback when the response is invalid or missing. (This template name can be customized via embed_obj.fallback_template_name.) Without a response, there won’t be any data to show in the normal/intended template. A fallback can provide more helpful output and a visual reference that something isn’t right.

Template tags/filters (requires lxml)–

resize_iframe is a template filter that caps the width of iframes since embedding an unexpectedly huge iframe into your layout might break the appearance. It only shrinks large iframes; it doesn’t alter iframes that are already the specified size (or smaller).

Common usage:
{{ object.response.render|resize_iframe:645|safe }}

In this example, if the render attribute contains code with iframes and the width of any or all of those iframes is larger than 645px, the iframes’ width will be changed to 645 and the height will scale smaller accordingly.


Content provider terms of service–

The service you are embedding content from may have usage guidelines and restrictions. Pay attention and follow these. It may be against their terms of service to rework or restyle the presentation or to use only pieces of the metadata. Changing or reusing things may also be disrespectful and disingenuous to the content creator. Respect the creator and respect the service.

Publishing content–

Embedded content is already published; it’s available from some other site. It’s how we use and integrate that external content into our own works that matters here. Armstrong is a platform for newsrooms and content publishers. The typical situation is one where reporters and editors write, draft, proof and publish. Content has eyes on it and doesn’t get published until it’s finished. AppsEmbeds is just the same. A general assumption is that some human is looking at the embed–maybe not the raw response data–but certainly the end result of how it looks (i.e. how a template renders it). If it looks wrong, it doesn’t get published.

It’s likely that someday you’ll come across a content provider whose responses don’t fit the expected form. It’s hard to account for these things but hopefully someone is looking at the content and will notice.

Custom API queries–

Many APIs provide customization for the responses they provide. They may allow you to specify maxwidth and maxheight, alignments for text or localization, callbacks, transparency modes or word length truncation. AppsEmbeds doesn’t do any of that primarily because it can’t make those assumptions. AppsEmbeds gets you the raw data in its default form whatever that may be and follows the “customize after” approach.

resize_iframe is an example of this. You may want a 200px iframe for a preview and an 800px iframe within an article body for the same embedded content. It wouldn’t do to set a maxwidth=200 on the API call, cache that and then be stuck for the larger size use case.

Ultimately, API use can be a finicky thing. The best course of action within the AppsEmbeds paradigm is to customize or create a backend and/or response class fitting the API you use and the parameters you may want to query with. Have a better idea or an awesome backend? Please make a Pull Request!

Different URLs to the same content–

There is currently no way to know if multiple URLs refer to the same content. These two YouTube links will make two separate Embed objects:


Development occurs on Github. Participation is welcome!

  • Found a bug? File it on Github Issues. Include as much detail as you can and make sure to list the specific component since we use a centralized, project-wide issue tracker.
  • Testing? pip install tox and run tox
  • Have code to submit? Fork the repo, consolidate your changes on a topic branch and create a pull request. The package provides tools for testing, coverage and South migration as well as making it very easy to run a full Django environment with this component’s settings.
  • Questions, need help, discussion? Use our Google Group mailing list.

State of Project

Armstrong is an open-source news platform that is freely available to any organization. It is the result of a collaboration between the Texas Tribune and The Center for Investigative Reporting and a grant from the John S. and James L. Knight Foundation. Armstrong is available as a complete bundle and as individual, stand-alone components.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Filename, size & hash SHA256 hash help File type Python version Upload date
armstrong.apps.embeds-0.9.tar.gz (27.0 kB) Copy SHA256 hash SHA256 Source None Sep 8, 2014

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page