Writing interactive bots

This guide is about writing and testing interactive bots. We assume familiarity with our guide for running bots.

On this page you'll find:

  • A step-by-step guide on how to set up a development environment for writing bots with all of our nice tooling to make it easy to write and test your work.
  • A guide on writing a bot.
  • A guide on adding a bot to Zulip.
  • A guide on testing a bot's output.
  • Documentation of the bot API.
  • Common problems when developing/running bots and their solutions.

Installing a development version of the Zulip bots package

  1. Clone the python-zulip-api repository:

    git clone https://github.com/zulip/python-zulip-api.git
  2. Navigate into your cloned repository:

    cd python-zulip-api
  3. Install all requirements in a Python virtualenv:

    python3 ./tools/provision
  4. Run the command provided in the final output of the provision process to enter the new virtualenv. The command will be of the form source .../activate.

  5. You should now see the name of your virtualenv preceding your prompt e.g., (zulip-api-py3-venv).

provision installs the zulip, zulip_bots, and zulip_botserver packages in developer mode. This enables you to modify these packages and then run your modified code without having to first re-install the packages or re-provision.

Writing a bot

The tutorial below explains the structure of a bot <my-bot>.py, which is the only file you need to create for a new bot. You can use this as boilerplate code for developing your own bot.

Every bot is built upon this structure:

class MyBotHandler(object):
    A docstring documenting this bot.

    def usage(self):
        return '''Your description of the bot'''

    def handle_message(self, message, bot_handler):
        # add your code here

handler_class = MyBotHandler
  • The class name (in this case MyBotHandler) can be defined by you and should match the name of your bot. To register your bot's class, adjust the last line handler_class = MyBotHandler to match your class name.

  • Every bot needs to implement the functions

    • usage(self)
    • handle_message(self, message, bot_handler)
  • These functions are documented in the next section.

Adding a bot to Zulip

Zulip's bot system resides in the python-zulip-api repository.

The structure of the bots ecosystem looks like the following:

    │   ├───bot1
    │   └───bot2
    │       │
    │       ├───bot2.py
    │       ├───bot2.conf
    │       ├───doc.md
    │       ├───requirements.txt
    │       ├───test_bot2.py
    │       ├───assets
    │       │   │
    │       │   └───pic.png
    │       ├───fixtures
    │       │   │
    │       │   └───test1.json
    │       └───libraries
    │           │
    │           └───lib1.py
    ├─── lib.py
    ├─── test_lib.py
    ├─── run.py
    └─── provision.py

Each subdirectory in bots contains a bot. When writing bots, try to use the structure outlined above as an orientation.

Testing a bot's output

If you just want to see how a bot reacts to a message, but don't want to set it up on a server, we have a little tool to help you out: zulip-bot-shell

Example invocations are below:

> zulip-bot-shell converter

Enter your message: "12 meter yard"
Response: 12.0 meter = 13.12336 yard

> zulip-bot-shell -b ~/followup.conf followup

Enter your message: "Task completed"
Response: stream: followup topic: foo_sender@zulip.com
          from foo_sender@zulip.com: Task completed

Note that the -b (aka --bot-config-file) argument is for an optional third party config file (e.g. ~/giphy.conf), which only applies to certain types of bots.


This section documents functions available to the bot and the structure of the bot's config file.

With this API, you can

  • intercept, view, and process messages sent by users on Zulip.
  • send out new messages as replies to the processed messages.

With this API, you cannot

  • modify an intercepted message (you have to send a new message).
  • send messages on behalf of or impersonate other users.
  • intercept direct messages (except for direct messages with the bot as an explicit recipient).



is called to retrieve information about the bot.


  • self - the instance the method is called on.

Return values

  • A string describing the bot's functionality

Example implementation

def usage(self):
    return '''
        This plugin will allow users to flag messages
        as being follow-up items.  Users should preface
        messages with "@followup".
        Before running this, make sure to create a channel
        called "followup" that your API user can send to.


handle_message(self, message, bot_handler)

handles user message.


  • self - the instance the method is called on.

  • message - a dictionary describing a Zulip message

  • bot_handler - used to interact with the server, e.g. to send a message

Return values


Example implementation

  def handle_message(self, message, bot_handler):
     original_content = message['content']
     original_sender = message['sender_email']
     new_content = original_content.replace('@followup',
                                            'from %s:' % (original_sender,))




will send a message as the bot user. Generally, this is less convenient than send_reply, but it offers additional flexibility about where the message is sent to.


  • message - a dictionary describing the message to be sent by the bot

Example implementation

    type='stream', # can be 'stream' or 'private'
    to=channel_name, # either the channel name or user's email
    subject=subject, # message subject
    content=message, # content of the sent message


bot_handler.send_reply(message, response)

will reply to the triggering message to the same place the original message was sent to, with the content of the reply being response.


  • message - Dictionary containing information on message to respond to (provided by handle_message).
  • response - Response message from the bot (string).



will edit the content of a previously sent message.


  • message - dictionary defining what message to edit and the new content


From zulip_bots/bots/incrementor/incrementor.py:

    message_id=self.message_id, # id of message to be updated
    content=str(self.number), # string with which to update message with


A common problem when writing an interactive bot is that you want to be able to store a bit of persistent state for the bot (e.g. for an RSVP bot, the RSVPs). For a sufficiently complex bot, you want need your own database, but for simpler bots, we offer a convenient way for bot code to persistently store data.

The interface for doing this is bot_handler.storage.

The data is stored in the Zulip Server's database. Each bot user has an independent storage quota available to it.

Performance considerations

Since each access to bot_handler.storage will involve a round-trip to the server, we recommend writing bots so that they do a single bot_handler.storage.get at the start of handle_message, and a single bot_handler.storage.put at the end to submit the state to the server. We plan to offer a context manager that takes care of this automatically.


bot_handler.storage.put(key, value)

will store the value value in the entry key.

  • key - a UTF-8 string
  • value - a UTF-8 string
bot_handler.storage.put("foo", "bar")  # set entry "foo" to "bar"



will retrieve the value for the entry key.

  • key - a UTF-8 string
bot_handler.storage.put("foo", "bar")
print(bot_handler.storage.get("foo"))  # print "bar"



will check if the entry key exists.

  • key - a UTF-8 string
bot_handler.storage.contains("foo")  # False
bot_handler.storage.put("foo", "bar")
bot_handler.storage.contains("foo")  # True

bot_handler.storage marshaling

By default, bot_handler.storage accepts any object for keys and values, as long as it is JSON-able. Internally, the object then gets converted to an UTF-8 string. You can specify custom data marshaling by setting the functions bot_handler.storage.marshal and bot_handler.storage.demarshal. These functions parse your data on every call to put and get, respectively.

Configuration file

  • key - the API key you created for the bot; this is how Zulip knows the request is from an authorized user.

  • email - the email address of the bot, e.g. some-bot@zulip.com

  • site - your development environment URL; if you are working on a development environment hosted on your computer, use localhost:9991

Writing tests for bots

Bots, like most software that you want to work, should have unit tests. In this section, we detail our framework for writing unit tests for bots. We require that bots in the main python-zulip-api repository include a reasonable set of unit tests, so that future developers can easily refactor them.

Unit tests for bots make heavy use of mocking. If you want to get comfortable with mocking, mocking strategies, etc. you should check out our mocking guide.

A simple example

Let's have a look at a simple test suite for the helloworld bot.

from zulip_bots.test_lib import StubBotTestCase

class TestHelpBot(StubBotTestCase):
    bot_name: str = "helloworld"

    def test_bot(self) -> None:
        dialog = [
            ('', 'beep boop'),
            ('help', 'beep boop'),
            ('foo', 'beep boop'),


The helloworld bot replies with "beep boop" to every message @-mentioning it. We want our test to verify that the bot actually does that.

Note that our helper method verify_dialog simulates the conversation for us, and we just need to set up a list of tuples with expected results.

The best way to learn about bot tests is to read all the existing tests in the bots subdirectories.

Testing your test

Once you have written a test suite, you want to verify that everything works as expected.

  • To test a bot in Zulip's bot directory: tools/test-bots <botname>

  • To run all bot tests: tools/test-bots

Advanced testing

This section shows advanced testing techniques for more complicated bots that have configuration files or interact with third-party APIs. The code for the bot testing library can be found here.

Testing bots with config files

Some bots, such as Giphy, support or require user configuration options to control how the bot works.

To test such a bot, you can use the following pattern:

with self.mock_config_info(dict(api_key=12345)):
    # self.verify_reply(...)

mock_config_info() replaces the actual step of reading configuration from the file system and gives your test "dummy data" instead.

Testing bots with internet access

Some bots, such as Giphy, depend on a third-party service, such as the Giphy web app, in order to work. Because we want our test suite to be reliable and not add load to these third-party APIs, tests for these services need to have "test fixtures": sample HTTP request/response pairs to be used by the tests. You can specify which one to use in your test code using the following helper method:

with self.mock_http_conversation('test_fixture_name'):
    # self.assert_bot_response(...)

mock_http_conversation(fixture_name) patches requests.get and returns the data specified in the file fixtures/{fixture_name}.json. Use the following JSON code as a skeleton for new fixtures:

  "request": {
    "api_url": "http://api.example.com/",
    "params": {
  "response": {
  "response-headers": {

For an example, check out the giphy bot.

Tip: You can use requestbin or a similar tool to capture payloads from the service your bot is interacting with.


Check out our bots to see examples of bot tests.

Common problems

  • I modified my bot's code, yet the changes don't seem to have an effect.

    • Ensure that you restarted the zulip-run-bot script.
  • My bot won't start

    • Ensure that your API config file is correct (download the config file from the server).
    • Ensure that you bot script is located in zulip_bots/bots/<my-bot>/
    • Are you using your own Zulip development server? Ensure that you run your bot outside the Vagrant environment.
    • Some bots require Python 3. Try switching to a Python 3 environment before running your bot.

Future direction

The long-term plan for this bot system is to allow the same ExternalBotHandler code to eventually be usable in several contexts:

  • Run directly using the Zulip call_on_each_message API, which is how the implementation above works. This is great for quick development with minimal setup.
  • Run in a simple Python webserver server, processing messages received from Zulip's outgoing webhooks integration.
  • For bots merged into the mainline Zulip codebase, enabled via a button in the Zulip web UI, with no code deployment effort required.