Quick Tour of Pyramid

Pyramid lets you start small and finish big. This Quick Tour of Pyramid is for those who want to evaluate Pyramid, whether you are new to Python web frameworks, or a pro in a hurry. For more detailed treatment of each topic, give the Quick Tutorial for Pyramid a try.


Once you have a standard Python environment setup, getting started with Pyramid is a breeze. Unfortunately "standard" is not so simple in Python. For this Quick Tour, it means: Python, a virtual environment (or virtualenv for Python 2.7), and setuptools.

As an example, for Python 3.3+ on Linux:

$ pyvenv env33
$ wget https://bitbucket.org/pypa/setuptools/raw/bootstrap/ez_setup.py -O - | env33/bin/python
$ env33/bin/easy_install "pyramid==1.5.7"

For Windows:

# Use your browser to download:
#   https://bitbucket.org/pypa/setuptools/raw/bootstrap/ez_setup.py
c:\> c:\Python33\python -m venv env33
c:\> env33\Scripts\python ez_setup.py
c:\> env33\Scripts\easy_install "pyramid==1.5.7"

Of course Pyramid runs fine on Python 2.6+, as do the examples in this Quick Tour. We're just showing Python 3 a little love (Pyramid had production support for Python 3 in October 2011).


Why easy_install and not pip? Some distributions on which Pyramid depends upon have optional C extensions for performance. pip cannot install some binary Python distributions. With easy_install, Windows users are able to obtain binary Python distributions, so they get the benefit of the C extensions without needing a C compiler. Also, there can be issues when pip and easy_install are used side-by-side in the same environment, so we chose to recommend easy_install for the sake of reducing the complexity of these instructions.

Hello World

Microframeworks have shown that learning starts best from a very small first step. Here's a tiny application in Pyramid:

from wsgiref.simple_server import make_server
from pyramid.config import Configurator
from pyramid.response import Response

def hello_world(request):
    return Response('<h1>Hello World!</h1>')

if __name__ == '__main__':
    config = Configurator()
    config.add_route('hello', '/')
    config.add_view(hello_world, route_name='hello')
    app = config.make_wsgi_app()
    server = make_server('', 6543, app)

This simple example is easy to run. Save this as app.py and run it:

$ python ./app.py

Next open http://localhost:6543/ in a browser, and you will see the Hello World! message.

New to Python web programming? If so, some lines in the module merit explanation:

  1. Line 10. The if __name__ == '__main__': is Python's way of saying "Start here when running from the command line".
  2. Lines 11-13. Use Pyramid's configurator to connect view code to a particular URL route.
  3. Lines 6-7. Implement the view code that generates the response.
  4. Lines 14-16. Publish a WSGI app using an HTTP server.

As shown in this example, the configurator plays a central role in Pyramid development. Building an application from loosely-coupled parts via Application Configuration is a central idea in Pyramid, one that we will revisit regurlarly in this Quick Tour.

Handling web requests and responses

Developing for the web means processing web requests. As this is a critical part of a web application, web developers need a robust, mature set of software for web requests.

Pyramid has always fit nicely into the existing world of Python web development (virtual environments, packaging, scaffolding, one of the first to embrace Python 3, etc.). Pyramid turned to the well-regarded WebOb Python library for request and response handling. In our example above, Pyramid hands hello_world a request that is based on WebOb.

Let's see some features of requests and responses in action:

def hello_world(request):
    # Some parameters from a request such as /?name=lisa
    url = request.url
    name = request.params.get('name', 'No Name Provided')

    body = 'URL %s with name: %s' % (url, name)
    return Response(

In this Pyramid view, we get the URL being visited from request.url. Also, if you visited http://localhost:6543/?name=alice in a browser, the name is included in the body of the response:

URL http://localhost:6543/?name=alice with name: alice

Finally, we set the response's content type and return the Response.


For the examples above, the hello_world function is a "view". In Pyramid, views are the primary way to accept web requests and return responses.

So far our examples place everything in one file:

  • the view function
  • its registration with the configurator
  • the route to map it to an URL
  • the WSGI application launcher

Let's move the views out to their own views.py module and change the app.py to scan that module, looking for decorators that set up the views.

First, our revised app.py:

from wsgiref.simple_server import make_server
from pyramid.config import Configurator

if __name__ == '__main__':
    config = Configurator()
    config.add_route('home', '/')
    config.add_route('hello', '/howdy')
    config.add_route('redirect', '/goto')
    config.add_route('exception', '/problem')
    app = config.make_wsgi_app()
    server = make_server('', 6543, app)

We added some more routes, but we also removed the view code. Our views and their registrations (via decorators) are now in a module views.py, which is scanned via config.scan('views').

We now have a views.py module that is focused on handling requests and responses:

from pyramid.httpexceptions import HTTPFound
from pyramid.response import Response
from pyramid.view import view_config

# First view, available at http://localhost:6543/
def home_view(request):
    return Response('<p>Visit <a href="/howdy?name=lisa">hello</a></p>')

# /howdy?name=alice which links to the next view
def hello_view(request):
    name = request.params.get('name', 'No Name')
    body = '<p>Hi %s, this <a href="/goto">redirects</a></p>'
    return Response(body % name)

# /goto which issues HTTP redirect to the last view
def redirect_view(request):
    return HTTPFound(location="/problem")

# /problem which causes a site error
def exception_view(request):
    raise Exception()

We have four views, each leading to the other. If you start at http://localhost:6543/, you get a response with a link to the next view. The hello_view (available at the URL /howdy) has a link to the redirect_view, which issues a redirect to the final view.

Earlier we saw config.add_view as one way to configure a view. This section introduces @view_config. Pyramid's configuration supports imperative configuration, such as the config.add_view in the previous example. You can also use declarative configuration, in which a Python decorator is placed on the line above the view. Both approaches result in the same final configuration, thus usually it is simply a matter of taste.


Writing web applications usually means sophisticated URL design. We just saw some Pyramid machinery for requests and views. Let's look at features that help in routing.

Above we saw the basics of routing URLs to views in Pyramid:

  • Your project's "setup" code registers a route name to be used when matching part of the URL
  • Elsewhere a view is configured to be called for that route name


Why do this twice? Other Python web frameworks let you create a route and associate it with a view in one step. As illustrated in Routes Need Relative Ordering, multiple routes might match the same URL pattern. Rather than provide ways to help guess, Pyramid lets you be explicit in ordering. Pyramid also gives facilities to avoid the problem.

What if we want part of the URL to be available as data in my view? This route declaration:

    config.add_route('hello', '/howdy/{first}/{last}')

With this, URLs such as /howdy/amy/smith will assign amy to first and smith to last. We can then use this data in our view:

def hello_world(request):
    body = '<h1>Hi %(first)s %(last)s!</h1>' % request.matchdict
    return Response(body)

request.matchdict contains values from the URL that match the "replacement patterns" (the curly braces) in the route declaration. This information can then be used in your view.


Ouch. We have been making our own Response and filling the response body with HTML. You usually won't embed an HTML string directly in Python, but instead, will use a templating language.

Pyramid doesn't mandate a particular database system, form library, etc. It encourages replaceability. This applies equally to templating, which is fortunate: developers have strong views about template languages. That said, the Pylons Project officially supports bindings for Chameleon, Jinja2, and Mako, so in this step, let's use Chameleon.

Let's add pyramid_chameleon, a Pyramid add-on which enables Chameleon as a renderer in our Pyramid applications:

$ easy_install pyramid_chameleon

With the package installed, we can include the template bindings into our configuration:


Now lets change our views.py file:

@view_config(route_name='hello', renderer='hello_world.pt')
def hello_world(request):
    return dict(name=request.matchdict['name'])

Ahh, that looks better. We have a view that is focused on Python code. Our @view_config decorator specifies a renderer that points to our template file. Our view then simply returns data which is then supplied to our template:

<!DOCTYPE html>
<html lang="en">
    <title>Quick Glance</title>
<h1>Hello ${name}</h1>

Since our view returned dict(name=request.matchdict['name']), we can use name as a variable in our template via ${name}.

Templating with jinja2

We just said Pyramid doesn't prefer one templating language over another. Time to prove it. Jinja2 is a popular templating system, modeled after Django's templates. Let's add pyramid_jinja2, a Pyramid add-on which enables Jinja2 as a renderer in our Pyramid applications:

$ easy_install pyramid_jinja2

With the package installed, we can include the template bindings into our configuration:


The only change in our view is to point the renderer at the .jinja2 file:

@view_config(route_name='hello', renderer='hello_world.jinja2')

Our Jinja2 template is very similar to our previous template:

<!DOCTYPE html>
<html lang="en">
    <title>Hello World</title>
<h1>Hello {{ name }}!</h1>

Pyramid's templating add-ons register a new kind of renderer into your application. The renderer registration maps to different kinds of filename extensions. In this case, changing the extension from .pt to .jinja2 passed the view response through the pyramid_jinja2 renderer.

Static assets

Of course the Web is more than just markup. You need static assets: CSS, JS, and images. Let's point our web app at a directory where Pyramid will serve some static assets. First another call to the configurator:

    config.add_static_view(name='static', path='static')

This tells our WSGI application to map requests under http://localhost:6543/static/ to files and directories inside a static directory alongside our Python module.

Next make a directory named static, and place app.css inside:

body {
    margin: 2em;
    font-family: sans-serif;

All we need to do now is point to it in the <head> of our Jinja2 template:

    <link rel="stylesheet" href="/static/app.css" />

This link presumes that our CSS is at a URL starting with /static/. What if the site is later moved under /somesite/static/? Or perhaps a web developer changes the arrangement on disk? Pyramid provides a helper to allow flexibility on URL generation:

    <link rel="stylesheet"

By using request.static_url to generate the full URL to the static assets, you both ensure you stay in sync with the configuration and gain refactoring flexibility later.

Returning JSON

Modern web apps are more than rendered HTML. Dynamic pages now use JavaScript to update the UI in the browser by requesting server data as JSON. Pyramid supports this with a JSON renderer:

@view_config(route_name='hello_json', renderer='json')
def hello_json(request):
    return [1, 2, 3]
    # End View 1

This wires up a view that returns some data through the JSON renderer, which calls Python's JSON support to serialize the data into JSON and set the appropriate HTTP headers.

View classes

So far our views have been simple, free-standing functions. Many times your views are related: different ways to look at or work on the same data, or a REST API that handles multiple operations. Grouping these together as a view class makes sense.

  • Group views
  • Centralize some repetitive defaults
  • Share some state and helpers

The following shows a "Hello World" example with three operations: view a form, save a change, or press the delete button:

# One route, at /howdy/amy, so don't repeat on each @view_config
class HelloWorldViews:
    def __init__(self, request):
        self.request = request
        # Our templates can now say {{ view.name }}
        self.name = request.matchdict['name']

    # Retrieving /howdy/amy the first time
    def hello_view(self):
        return dict()

    # Posting to /howdy/amy via the "Edit" submit button
    @view_config(request_param='form.edit', renderer='edit.jinja2')
    def edit_view(self):
        return dict()

    # Posting to /howdy/amy via the "Delete" submit button
    @view_config(request_param='form.delete', renderer='delete.jinja2')
    def delete_view(self):
        return dict()

As you can see, the three views are logically grouped together. Specifically:

  • The first view is returned when you go to /howdy/amy. This URL is mapped to the hello route that we centrally set using the optional @view_defaults.
  • The second view is returned when the form data contains a field with form.edit, such as clicking on <input type="submit" name="form.edit" value="Save">. This rule is specified in the @view_config for that view.
  • The third view is returned when clicking on a button such as <input type="submit" name="form.delete" value="Delete">.

Only one route is needed, stated in one place atop the view class. Also, the assignment of name is done in the __init__ function. Our templates can then use {{ view.name }}.

Pyramid view classes, combined with built-in and custom predicates, have much more to offer:

  • All the same view configuration parameters as function views
  • One route leading to multiple views, based on information in the request or data such as request_param, request_method, accept, header, xhr, containment, and custom_predicates

Quick project startup with scaffolds

So far we have done all of our Quick Tour as a single Python file. No Python packages, no structure. Most Pyramid projects, though, aren't developed this way.

To ease the process of getting started, Pyramid provides scaffolds that generate sample projects from templates in Pyramid and Pyramid add-ons. Pyramid's pcreate command can list the available scaffolds:

$ pcreate --list
Available scaffolds:
  alchemy:                 Pyramid SQLAlchemy project using url dispatch
  pyramid_jinja2_starter:  pyramid jinja2 starter project
  starter:                 Pyramid starter project
  zodb:                    Pyramid ZODB project using traversal

The pyramid_jinja2 add-on gave us a scaffold that we can use. From the parent directory of where we want our Python package to be generated, let's use that scaffold to make our project:

$ pcreate --scaffold pyramid_jinja2_starter hello_world

We next use the normal Python command to set up our package for development:

$ cd hello_world
$ python ./setup.py develop

We are moving in the direction of a full-featured Pyramid project, with a proper setup for Python standards (packaging) and Pyramid configuration. This includes a new way of running your application:

$ pserve development.ini

Let's look at pserve and configuration in more depth.

Application running with pserve

Prior to scaffolds, our project mixed a number of operational details into our code. Why should my main code care which HTTP server I want and what port number to run on?

pserve is Pyramid's application runner, separating operational details from your code. When you install Pyramid, a small command program called pserve is written to your bin directory. This program is an executable Python module. It's very small, getting most of its brains via import.

You can run pserve with --help to see some of its options. Doing so reveals that you can ask pserve to watch your development files and reload the server when they change:

$ pserve development.ini --reload

The pserve command has a number of other options and operations. Most of the work, though, comes from your project's wiring, as expressed in the configuration file you supply to pserve. Let's take a look at this configuration file.

See also

See also: What Is This pserve Thing

Configuration with .ini files

Earlier in Quick Tour we first met Pyramid's configuration system. At that point we did all configuration in Python code. For example, the port number chosen for our HTTP server was right there in Python code. Our scaffold has moved this decision and more into the development.ini file:

# Start Includes
pyramid.includes = pyramid_debugtoolbar
# End Includes
use = egg:hello_world
reload_templates = true
debug_authorization = false
debug_notfound = false
debug_routematch = false
debug_templates = true
default_locale_name = en
jinja2.directories = hello_world:templates

pipeline =

use = egg:pyramid#wsgiref
host =
port = 6543

# Begin logging configuration

# Start Sphinx Include
keys = root, hello_world

level = DEBUG
handlers =
qualname = hello_world
# End Sphinx Include

keys = console

keys = generic

level = INFO
handlers = console

class = StreamHandler
args = (sys.stderr,)
level = NOTSET
formatter = generic

format = %(asctime)s %(levelname)-5.5s [%(name)s][%(threadName)s] %(message)s

# End logging configuration

Let's take a quick high-level look. First the .ini file is divided into sections:

  • [app:hello_world] configures our WSGI app
  • [pipeline:main] sets up our WSGI "pipeline"
  • [server:main] holds our WSGI server settings
  • Various sections afterwards configure our Python logging system

We have a few decisions made for us in this configuration:

  1. Choice of web server: use = egg:pyramid#wsgiref tells pserve to use the wsgiref server that is wrapped in the Pyramid package.
  2. Port number: port = 6543 tells wsgiref to listen on port 6543.
  3. WSGI app: What package has our WSGI application in it? use = egg:hello_world in the app section tells the configuration what application to load.
  4. Easier development by automatic template reloading: In development mode, you shouldn't have to restart the server when editing a Jinja2 template. reload_templates = true sets this policy, which might be different in production.

Additionally the development.ini generated by this scaffold wired up Python's standard logging. We'll now see in the console, for example, a log on every request that comes in, as well as traceback information.

Easier development with debugtoolbar

As we introduce the basics, we also want to show how to be productive in development and debugging. For example, we just discussed template reloading and earlier we showed --reload for application reloading.

pyramid_debugtoolbar is a popular Pyramid add-on which makes several tools available in your browser. Adding it to your project illustrates several points about configuration.

First change your setup.py to say:

requires = ['pyramid>=1.0.2', 'pyramid_jinja2', 'pyramid_debugtoolbar']

...and rerun your setup:

$ python ./setup.py develop

The Python package pyramid_debugtoolbar is now installed into our environment. The package is a Pyramid add-on, which means we need to include its configuration into our web application. We could do this with imperative configuration, as we did above for the pyramid_jinja2 add-on:


Now that we have a configuration file, we can use the pyramid.includes facility and place this in our development.ini instead:

pyramid.includes = pyramid_debugtoolbar

You'll now see an attractive (and collapsible) menu in the right of your browser, providing introspective access to debugging information. Even better, if your web application generates an error, you will see a nice traceback on the screen. When you want to disable this toolbar, there's no need to change code: you can remove it from pyramid.includes in the relevant .ini configuration file.

Unit tests and nose

Yikes! We got this far and we haven't yet discussed tests. This is particularly egregious, as Pyramid has had a deep commitment to full test coverage since before its release.

Our pyramid_jinja2_starter scaffold generated a tests.py module with one unit test in it. To run it, let's install the handy nose test runner by editing setup.py. While we're at it, we'll throw in the coverage tool which yells at us for code that isn't tested:

      # Some lines removed...
          'testing': ['nose', 'coverage'],

We changed setup.py which means we need to rerun python ./setup.py develop. We can now run all our tests:

$ nosetests hello_world/tests.py
Name                  Stmts   Miss  Cover   Missing
hello_world             12      8    33%   11-23
hello_world.models       5      1    80%   8
hello_world.tests       14      0   100%
hello_world.views        4      0   100%
TOTAL                    35      9    74%
Ran 1 test in 0.931s


Our unit test passed. What did our test look like?

import unittest
from pyramid import testing
from pyramid.i18n import TranslationStringFactory

_ = TranslationStringFactory('hello_world')

class ViewTests(unittest.TestCase):

    def setUp(self):
    def tearDown(self):

    def test_my_view(self):
        from hello_world.views import my_view
        request = testing.DummyRequest()
        response = my_view(request)
        self.assertEqual(response['project'], 'hello_world')

Pyramid supplies helpers for test writing, which we use in the test setup and teardown. Our one test imports the view, makes a dummy request, and sees if the view returns what we expected.


It's important to know what is going on inside our web application. In development we might need to collect some output. In production we might need to detect situations when other people use the site. We need logging.

Fortunately Pyramid uses the normal Python approach to logging. The scaffold generated in your development.ini has a number of lines that configure the logging for you to some reasonable defaults. You then see messages sent by Pyramid (for example, when a new request comes in).

Maybe you would like to log messages in your code? In your Python module, import and set up the logging:

import logging
log = logging.getLogger(__name__)

You can now, in your code, log messages:

    log.debug('Some Message')

This will log Some Message at a debug log level to the application-configured logger in your development.ini. What controls that? These sections in the configuration file:

keys = root, hello_world

level = DEBUG
handlers =
qualname = hello_world

Our application, a package named hello_world, is set up as a logger and configured to log messages at a DEBUG or higher level. When you visit http://localhost:6543, your console will now show:

2013-08-09 10:42:42,968 DEBUG [hello_world.views][MainThread] Some Message

See also

See also: Quick Tutorial Logging and Logging


When people use your web application, they frequently perform a task that requires semi-permanent data to be saved. For example, a shopping cart. This is called a session.

Pyramid has basic built-in support for sessions. Third party packages such as pyramid_redis_sessions provide richer session support. Or you can create your own custom sessioning engine. Let's take a look at the built-in sessioning support. In our __init__.py we first import the kind of sessioning we want:

from pyramid.session import SignedCookieSessionFactory


As noted in the session docs, this example implementation is not intended for use in settings with security implications.

Now make a "factory" and pass it to the configurator's session_factory argument:

    my_session_factory = SignedCookieSessionFactory('itsaseekreet')
    config = Configurator(root_factory=get_root, settings=settings,

Pyramid's request object now has a session attribute that we can use in our view code:

    session = request.session
    if 'counter' in session:
        session['counter'] += 1
        session['counter'] = 0

With this, each reload will increase the counter displayed in our Jinja2 template:

          <p>Counter: {{ request.session.counter }}</p>


Web applications mean data. Data means databases. Frequently SQL databases. SQL databases frequently mean an "ORM" (object-relational mapper.) In Python, ORM usually leads to the mega-quality SQLAlchemy, a Python package that greatly eases working with databases.

Pyramid and SQLAlchemy are great friends. That friendship includes a scaffold!

$ pcreate --scaffold alchemy sqla_demo
$ cd sqla_demo
$ python setup.py develop

We now have a working sample SQLAlchemy application with all dependencies installed. The sample project provides a console script to initialize a SQLite database with tables. Let's run it and then start the application:

$ initialize_sqla_demo_db development.ini
$ pserve development.ini

The ORM eases the mapping of database structures into a programming language. SQLAlchemy uses "models" for this mapping. The scaffold generated a sample model:

class MyModel(Base):
    __tablename__ = 'models'
    id = Column(Integer, primary_key=True)
    name = Column(Text, unique=True)
    value = Column(Integer)

    def __init__(self, name, value):
        self.name = name
        self.value = value

View code, which mediates the logic between web requests and the rest of the system, can then easily get at the data thanks to SQLAlchemy:

        one = DBSession.query(MyModel).filter(MyModel.name == 'one').first()


Developers have lots of opinions about web forms, and thus there are many form libraries for Python. Pyramid doesn't directly bundle a form library, but Deform is a popular choice for forms, along with its related Colander schema system.

As an example, imagine we want a form that edits a wiki page. The form should have two fields on it, one of them a required title and the other a rich text editor for the body. With Deform we can express this as a Colander schema:

class WikiPage(colander.MappingSchema):
    title = colander.SchemaNode(colander.String())
    body = colander.SchemaNode(

With this in place, we can render the HTML for a form, perhaps with form data from an existing page:

form = self.wiki_form.render()

We'd like to handle form submission, validation, and saving:

# Get the form data that was posted
controls = self.request.POST.items()
    # Validate and either raise a validation error
    # or return deserialized data from widgets
    appstruct = wiki_form.validate(controls)
except deform.ValidationFailure as e:
    # Bail out and render form with errors
    return dict(title=title, page=page, form=e.render())

# Change the content and redirect to the view
page['title'] = appstruct['title']
page['body'] = appstruct['body']

Deform and Colander provide a very flexible combination for forms, widgets, schemas, and validation. Recent versions of Deform also include a retail mode for gaining Deform features on custom forms.

Also the deform_bootstrap Pyramid add-on restyles the stock Deform widgets using attractive CSS from Twitter Bootstrap and more powerful widgets from Chosen.


This Quick Tour covered a little about a lot. We introduced a long list of concepts in Pyramid, many of which are expanded on more fully in the Pyramid developer docs.