Step-by-Step Guide: How to Build Your Own Chatbot with the ChatGPT API

In this blog post, we will be discussing how to build your own chat bot using the ChatGPT API. It’s worth mentioning that we will be using the OpenAI APIs directly and not the Azure OpenAI APIs, and the code will be written in Python. A crucial aspect of creating a chat bot is maintaining context in the conversation, which we will achieve by storing and sending previous messages to the API at each request. If you are just starting with AI and chat bots, this post will guide you through the step-by-step process of building your own simple chat bot using the ChatGPT API.

Python setup

Ensure Python is installed. I am using version 3.10.8. For editing code, I am using Visual Studio code as the editor. For the text-based chat bot, you will need the following Python packages:

  • openai: make sure the version is 0.27.0 or higher; earlier versions do not support the ChatCompletion APIs
  • tiktoken: a library to count the number of tokens of your chat bot messages

Install the above packages with your package manager. For example: pip install openai.

All code can be found on GitHub.

Getting an account at OpenAI

We will write a text-based chat bot that asks for user input indefinitely. The first thing you need to do is sign up for API access at Access is not free but for personal use, while writing and testing the chat bot, the price will be very low. Here is a screenshot from my account:

Oh no, $0.13 dollars

When you have your account, generate an API key from Click the Create new secret key button and store the key somewhere.

Writing the bot

Now create a new Python file called and add the following lines:

import os
import openai
import tiktoken

openai.api_key = os.getenv("OPENAI_KEY")

We already discussed the openai and tiktoken libraries. We will also use the builtin os library to read environment variables.

In the last line, we read the environment variable OPENAI_KEY. If you use Linux, in your shell, use the following command to store the OpenAI key in an environment variable: export OPENAI_KEY=your-OpenAI-key. We use this approach to avoid storing the API key in your code and accidentally uploading it to GitHub.

To implement the core chat functionality, we will use a Python class. I was following a Udemy course about ChatGPT and it used a similar approach, which I liked. By the way, I can highly recommend that course. Check it out here.

Let’s start with the class constructor:

class ChatBot:

    def __init__(self, message):
        self.messages = [
            { "role": "system", "content": message }

In the constructor, we define a messages list and set the first item in that list to a configurable dictionary: { "role": "system", "content": message }. In the ChatGPT API calls, the messages list provides context to the API because it contains all the previous messages. With this initial system message, we can instruct the API to behave in a certain way. For example, later in the code, you will find this code to create an instance of the ChatBot class:

bot = ChatBot("You are an assistant that always answers correctly. If not sure, say 'I don't know'.")

But you could also do:

bot = ChatBot("You are an assistant that always answers wrongly.Always contradict the user")

In practice, ChatGPT does not follow the system instruction to strongly. User messages are more important. So it could be that, after some back and forth, the answers will not follow the system instruction anymore.

Let’s continue with another method in the class, the chat method:

def chat(self):
        prompt = input("You: ")
            { "role": "user", "content": prompt}
        response = openai.ChatCompletion.create(
            messages = self.messages,
            temperature = 0.8
        answer = response.choices[0]['message']['content']
           { "role": "assistant", "content": answer} 

        tokens = self.num_tokens_from_messages(self.messages)
        print(f"Total tokens: {tokens}")

        if tokens > 4000:
            print("WARNING: Number of tokens exceeds 4000. Truncating messages.")
            self.messages = self.messages[2:]

The chat method is where the action happens. It does the following:

  • It prompts the user to enter some input.
  • The user’s input is stored in a dictionary as a message with a “user” role and appended to a list of messages called self.messages. If this is the first input, we now have two messages in the list, a system message and a user message.
  • It then creates a response using OpenAI’s gpt-3.5-turbo model, passing in the self.messages list and a temperature of 0.8 as parameters. We use the ChatCompletion API versus the Completion API that you use with other models such as text-davinci-003.
  • The generated response is stored in a variable named answer. The full response contains a lot of information. We are only interested in the first response (there is only one) and grab the content.
  • The answer is printed to the console.
  • The answer is also added to the self.messages list as a message with an “assistant” role. If this is the first input, we now have three messages in the list: a system message, the first user message (the input) and the assistant’s response.
  • The total number of tokens in the self.messages list is computed using a separate function called num_tokens_from_messages() and printed to the console.
  • If the number of tokens exceeds 4000, a warning message is printed and the self.messages list is truncated to remove the first two messages. We will talk about these tokens later.

It’s important to realize we are using the Chat completions here. You can find more information about Chat completions here.

If you did not quite get how the text response gets extracted, here is an example of a full response from the Chat completion API:

 'id': 'chatcmpl-6p9XYPYSTTRi0xEviKjjilqrWU2Ve',
 'object': 'chat.completion',
 'created': 1677649420,
 'model': 'gpt-3.5-turbo',
 'usage': {'prompt_tokens': 56, 'completion_tokens': 31, 'total_tokens': 87},
 'choices': [
    'message': {
      'role': 'assistant',
      'content': 'The 2020 World Series was played in Arlington, Texas at the Globe Life Field, which was the new home stadium for the Texas Rangers.'},
    'finish_reason': 'stop',
    'index': 0

The response is indeed in choices[0][‘message’][‘content’].

To make this rudimentary chat bot work, we will repeatedly call the chat method like so:

bot = ChatBot("You are an assistant that always answers correctly. If not sure, say 'I don't know'.")
    while True:

Every time you input a question, the API answers and both the question and answer is added to the messages list. Of course, that makes the messages list grow larger and larger, up to a point where it gets to large. The question is: “What is too large?”. Let’s answer that in the next section.

Counting tokens

A language model does not work with text as humans do. Instead, they use tokens. It’s not important how this exactly works but it is important to know that you get billed based on these tokens. You pay per token.

In addition, the model we use here (gpt-3.5-turbo) has a maximum limit of 4096 tokens. This might change in the future. With our code, we cannot keep adding messages to the messages list because, eventually, we will pass the limit and the API call will fail.

To have an idea about the tokens in our messages list, we have this function:

def num_tokens_from_messages(self, messages, model="gpt-3.5-turbo"):
            encoding = tiktoken.encoding_for_model(model)
        except KeyError:
            encoding = tiktoken.get_encoding("cl100k_base")
        if model == "gpt-3.5-turbo":  # note: future models may deviate from this
            num_tokens = 0
            for message in messages:
                num_tokens += 4  # every message follows <im_start>{role/name}\n{content}<im_end>\n
                for key, value in message.items():
                    num_tokens += len(encoding.encode(value))
                    if key == "name":  # if there's a name, the role is omitted
                        num_tokens += -1  # role is always required and always 1 token
            num_tokens += 2  # every reply is primed with <im_start>assistant
            return num_tokens
            raise NotImplementedError(f"""num_tokens_from_messages() is not presently implemented for model {model}.""")

The above function comes from the OpenAI cookbook on GitHub. In my code, the function is used to count tokens in the messages list and, if the number of tokens is above a certain limit, we remove the first two messages from the list. The code also prints the tokens so you now how many you will be sending to the API.

The function contains references to <im_start> and <im_end>. This is ChatML and is discussed here. Because you use the ChatCompletion API, you do not have to worry about this. You just use the messages list and the API will transform it all to ChatML. But when you count tokens, ChatML needs to be taken into account for the total token count.

Note that Microsoft examples for Azure OpenAI, do use ChatML in the prompt, in combination with the default Completion APIs. See Microsoft Learn for more information. You will quickly see that using the ChatCompletion API with the messages list is much simpler.

To see, and download, the full code, see GitHub.

Running the code

To run the code, just run On my system, I need to use python3 I set the system message to You are an assistant that always answers wrongly. Contradict the user. 😀

Here’s an example conversation:

Although, at the start, the responses follow the system message, the assistant starts to correct itself and answers correctly. As stated, user messages eventually carry more weight.


In this post, we discussed how to build a chat bot using the ChatGPT API and Python. We went through the setup process, created an OpenAI account, and wrote the chat bot code using the OpenAI API. The bot used the ChatCompletion API and maintained context in the conversation by storing and sending previous messages to the API at each request. We also discussed counting tokens and truncating the message list to avoid exceeding the maximum token limit for the model. The full code is available on GitHub, and we provided an example conversation between the bot and the user. The post aimed to guide both beginning developers and beginners in AI and chat bot development through the step-by-step process of building their chat bot using the ChatGPT API and keep it as simple as possible.

Hope you liked it!

A quick look at Azure App Configuration and the Python Provider

When developing an application, it is highly likely that it needs to be configured with all sorts of settings. A simple list of key/value pairs is usually all you need. Some of the values can be read by anyone (e.g., a public URL) while some values should be treated as secrets (e.g., a connection string).

Azure App Configuration is a service to centrally manage these settings in addition to feature flags. In this post, we will look at storing and retrieving application settings and keeping feature flags for another time. I will also say App Config instead of App Configuration to save some keystrokes. 😉

We will do the following:

  • Retrieve key-value pairs for multiple applications and environments from one App Config instance
  • Use Key Vault references in App Config and retrieve these from Key Vault directly
  • Use the Python provider client to retrieve key-value pairs and store them in a Python dictionary

Why use App Configuration at all?

App Configuration helps by providing a fully managed service to store configuration settings for your applications separately from your code. Storing configuration separate from code is a best practice that most developers should follow.

Although you could store configuration values in files, using a service like App Config provides some standardization within or across developer teams.

Some developers store both configuration values and secrets in Key Vault. Although that works, App Config is way more flexible in organizing the settings and retrieving lists of settings with key and label filters. If you need to work with more than a few settings, I would recommend using a combination of App Config and Key Vault.

In what follows, I will show how we store settings for multiple applications and environments in the same App Config instance. Some of these settings will be Key Vault references.

Read before continuing to know more about App Config.

Provisioning App Config

Provisioning App Configuration is very easy from the portal or the Azure CLI. With the Azure CLI, use the following commands to create a resource group and an App Configuration instance in that group:

az group create -n RESOURCEGROUP -l LOCATION

After deployment, we can check the portal and navigate to Configuration Explorer.

App Configuration in the Azure Portal

In Configuration Explorer, you can add the configuration values for your apps. They are just key/value pairs but they can be further enriched with labels, content types, and tags.

Note that there is a Free and a Standard tier of App Config. See for more information. In production, you should use the Standard tier.

Storing configuration and secrets for multiple apps and environments

To store configuration values for multiple applications, you will have to identify the application in the key. App Configuration, oddly, has no knowledge of applications. For example, a key could be app1:setting1. You decide on the separator between the app name (app1 here) and its setting (setting1). In your code, you can easily query all settings for your app with a key filter (e.g. “app1:”. I will show an example of using a key filter later with the Python provider.

If you want to have different values for a key per environment (like dev, prd, etc…), you can add a label for each environment. To retrieve all settings for an environment, you can use a label filter. I will show an example of using a label filter later.

Suppose you want to use app1:setting1 in two environments: dev and prd. How do you create the key-value pairs? One way is to use the Azure CLI. You can also create them with the portal or from Python, C#, etc… With the CLI:

az appconfig kv set --name APPCONFIGNAME  --key app1:setting1 --value "value1" --label dev

APPCONFIG name is the name of your App Config instance. Just the name, not the full URL. For the prd environment:

az appconfig kv set --name APPCONFIGNAME  --key app1:setting1 --value "value2" --label prd

In Configuration Explorer, you will now see:

app1:setting1 for two environments (via labels)

For more examples of using the Azure CLI, see

In addition to these plain key-value pairs, you can also create Key Vault references. Let’s create one from the portal. In Configuration Explorer, click + Create and select Key Vault reference. You will get the following UI that allows you to create the reference. Make sure you have a Key Vault with a secret called dev-mysecret if you want to follow along. Below, set the label to dev. I forgot that in the screenshot below:

Creating a Key Vault Reference

Above, I am using the same naming convention for the key in App Config: app1:mysecret. Notice though that the secret I am referencing in Key Vault contains the environment and a dash (-) before the actual secret name. If you use one Key Vault per app instead of a Key Vault per app and environment, you will have to identify the environment in the secret name in some way.

After creating the reference, you will see the following in Configuration explorer:

Configuration explorer with one Key Vault reference

Note that the Key Vault reference has a content type. The content type is application/;charset=utf-8. You can use the content type in your code to know if the key contains a reference to a Key Vault secret. That reference will be something like You can then use the Python SDK for Azure Key Vault to retrieve the secret from your code. Azure App Config will not do that for you.

You can use content types in other ways as well. For example, you could store a link to a storage account blob and use a content type that informs your code it needs to retrieve the blob from the account. Of course, you will need to write code to retrieve the blob. App Config only contains the reference.

Reading settings

There are many ways to read settings from App Config. If you need them in an Azure Pipeline, for instance, you can use the Azure App Configuration task to pull keys and values from App Config and set them as Azure pipeline variables.

If you deploy your app to Kubernetes and you do not want to read the settings from your code, you can integrate App Configuration with Helm. See for more information.

In most cases though, you will want to read the settings directly from your code. There is an SDK for several languages, including Python. The SDK has all the functionality you need to read and write settings.

Next to the Python SDK, there is also a Python provider which is optimized to read settings from App Config and store them in a Python dictionary. The provider has several options to automatically trim app names from keys and to automatically retrieve a secret from Key Vault if the setting in App Config is a Key Vault reference.

To authenticate to App Config, the default is access keys with a connection string. You can find the connection string in the Portal:

App Config Connection string for read/write or just read

You can also use Azure AD (it’s always enabled) and disable access keys. In this example, I will use a connection string to start with:

Before we connect and retrieve the values, ensure you install the provider first:

pip install azure-appconfiguration-provider

Above, use pip or pip3 depending on your installation of Python.

In your code, ensure the proper imports:

from azure.appconfiguration.provider import (
from azure.keyvault.secrets import SecretClient
from azure.identity import DefaultAzureCredential

To authenticate to Azure Key Vault with Azure AD, we can use DefaultAzureCredential():

    CREDENTIAL = DefaultAzureCredential(exclude_visual_studio_code_credential=True)
except Exception as ex:
    print(f"error setting credentials: {ex}")

Note: on my machine, I had an issue with the VS Code credential feature so I turned that off.

Next, use a SettingSelector from the provider to provide a key filter and label filter. I want to retrieve key-value pairs for an app called app1 and an environment called dev:

app = 'app1'
env = 'dev'
selects = {SettingSelector(key_filter=f"{app}:*", label_filter=env)}

Next, when I retrieve the key-value pairs, I want to strip app1: from the keys:

trimmed_key_prefixes = {f"{app}:"}

In addition, I want the provider to automatically go to Key Vault and retrieve the secret:

key_vault_options = AzureAppConfigurationKeyVaultOptions(secret_resolver=retrieve_secret)

retrieve_secret refers to a function you need to write to retrieve the secret and add custom logic. There are other options as well.

def retrieve_secret(uri):
        # uri is in format: https://<keyvaultname><secretname>
        # retrieve key vault uri and secret name from uri
        vault_uri = "https://" + uri.split('/')[2]
        secret_name = uri.split('/')[-1]
        print(f"Retrieving secret {secret_name} from {vault_uri}...")
        # retrieve the secret from Key Vault; CREDENTIAL was set globally
        secret_client = SecretClient(vault_url=vault_uri, credential=CREDENTIAL)
        # get secret value from Key Vault
        secret_value = secret_client.get_secret(secret_name).value
    except Exception as ex:
        print(f"retrieving secret: {ex}", 1)

    return secret_value

Now that we have all the options, we can retrieve the key-value pairs.

connection_string = 'YOURCONNSTR'
app_config = AzureAppConfigurationProvider.load(
    connection_string=connection_string, selects=selects, key_vault_options=key_vault_options, 


Now we have a Python dictionary app_config with all key-value pairs for app1 and environment dev. The key-value pairs are a mix of plain values from App Config and Key Vault.

You can now use this dictionary in your app in whatever way you like.

If you would like to use the same CREDENTIAL to connect to App Config, you can also use:

endpoint = '' # no https://
app_config = AzureAppConfigurationProvider.load(
    endpoint=endpoint, credential=CREDENTIAL, selects=selects, key_vault_options=key_vault_options, 

Ensure the credential you use has the App Configuration Data Reader role to read the key-value pairs.

Here’s all the code in a gist: Ensure you have the key-value pairs as above and provide the connection string to the connection_string variable.


In this post, we showed how to retrieve key-value pairs with the Python provider from one App Config instance for multiple applications and environments.

The application is stored as a prefix in the key (app1:). The environment is a label (e.g., dev), allowing us to have the same setting with different values per environment.

Some keys can contain a reference to Key Vault to allow your application to retrieve secrets from Key Vault as well. I like this approach to have a list of all settings for an app and environment, where the value of the key can be an actual value or a reference to some other entity like a secret, a blob, or anything else.

Writing a Kubernetes operator with Kopf

In today’s post, we will write a simple operator with Kopf, which is a Python framework created by Zalando. A Kubernetes operator is a piece of software, running in Kubernetes, that does something application specific. To see some examples of what operators are used for, check out

Our operator will do something simple in order to easily grasp how it works:

  • the operator will create a deployment that runs nginx
  • nginx will serve a static website based on a git repository that you specify; we will use an init container to grab the website from git and store it in a volume
  • you can control the number of instances via a replicas parameter

That’s great but how will the operator know when it has to do something, like creating or updating resources? We will use custom resources for that. Read on to learn more…

Note: source files are on GitHub

Custom Resource Definition (CRD)

Kubernetes allows you to define your own resources. We will create a resource of type (kind) DemoWeb. The CRD is created with the YAML below:

# A simple CRD to deploy a demo website from a git repo
kind: CustomResourceDefinition
  scope: Namespaced
    - name: v1
      served: true
      storage: true
    kind: DemoWeb
    plural: demowebs
    singular: demoweb
      - dweb
    - name: Replicas
      type: string
      priority: 0
      JSONPath: .spec.replicas
      description: Amount of replicas
    - name: GitRepo
      type: string
      priority: 0
      JSONPath: .spec.gitrepo
      description: Git repository with web content

For more information (and there is a lot) about CRDs, see the documentation.

Once you create the above resource with kubectl apply (or create), you can create a custom resource based on the definition:

kind: DemoWeb
  name: demoweb1
  replicas: 2
  gitrepo: ""

Note that we specified our own API and version in the CRD ( and that we set the kind to DemoWeb. In the additionalPrinterColumns, we defined some properties that can be set in the spec that will also be printed on screen. When you list resources of kind DemoWeb, you will the see replicas and gitrepo columns:

Custom resources based on the DemoWeb CRD

Of course, creating the CRD and the custom resources is not enough. To actually create the nginx deployment when the custom resource is created, we need to write and run the operator.

Writing the operator

I wrote the operator on a Mac with Python 3.7.6 (64-bit). On Windows, for best results, make sure you use Miniconda instead of Python from the Windows Store. First install Kopf and the Kubernetes package:

pip3 install kopf kubernetes

Verify you can run kopf:

Running kopf

Let’s write the operator. You can find it in full here. Here’s the first part:

Naturally, we import kopf and other necessary packages. As noted before, kopf and kubernetes will have to be installed with pip. Next, we define a handler that runs whenever a resource of our custom type is spotted by the operator (with the @kopf.on.create decorator). The handler has two parameters:

  • spec object: allows us to retrieve our custom properties with spec.get (e.g. spec.get(‘replicas’, 1) – the second parameter is the default value)
  • **kwargs: a dictionary with lots of extra values we can use; we use it to retrieve the name of our custom resource (e.g. demoweb1); we can use that name to derive the name of our deployment and to set labels for our pods

Note: instead of using **kwargs to retrieve the name, you can also define an extra name parameter in the handler like so: def create_fn(spec, name, **kwargs); see the docs for more information

Our deployment is just yaml stored in the doc variable with some help from the Python yaml package. We use spec.get and the name variable to customise it.

After the doc variable, the following code completes the event handler:

The rest of the operator

With kopf.adopt, we make sure the deployment we create is a child of our custom resource. When we delete the custom resource, its children are also deleted.

Next, we simply use the kubernetes client to create a deployment via the apps/v1 api. The method create_namespaced_deployment takes two required parameters: the namespace and the deployment specification. Note there is only minimal error checking here. There is much more you can do with regards to error checking, retries, etc…

Now we can run the operator with:

kopf run

You can perfectly run this on your local workstation if you have a working kube config pointing at a running cluster with the CRD installed. Kopf will automatically use that for authentication:

Running the operator on your workstation

Running the operator in your cluster

To run the operator in your cluster, create a Dockerfile that produces an image with Python, kopf, kubernetes and your operator in Python. In my case:

FROM python:3.7
RUN mkdir /src
ADD /src
RUN pip install kopf
RUN pip install kubernetes
CMD kopf run /src/ --verbose

We added the verbose parameter for extra logging. Next, run the following commands to build and push the image (example with my image name):

docker build -t gbaeke/kopf-demoweb .
docker push gbaeke/kopf-demoweb

Now you can deploy the operator to the cluster:

apiVersion: apps/v1
kind: Deployment
  name: demowebs-operator
  replicas: 1
    type: Recreate
      application: demowebs-operator
        application: demowebs-operator
      serviceAccountName: demowebs-account
      - name: demowebs
        image: gbaeke/kopf-demoweb

The above is just a regular deployment but the serviceAccountName is extremely important. It gives kopf and your operator the required access rights to create the deployment is the target namespace. Check out the documentation to find out more about the creation of the service account and the required roles. Note that you should only run one instance of the operator!

Once the operator is deployed, you will see it running as a normal pod:

The operator is running

To see what is going on, check the logs. Let’s show them with octant:

Your operator logs

At the bottom, you see what happens when a creation event is detected for a resource of type DemoWeb. The spec is shown with the git repository and the number on replicas.

Now you can create resources of kind DemoWeb and see what happens. If you have your own git repository with some HTML in it, try to use that. Otherwise, just use mine at


Writing an operator is easy to do with the Kopf framework. Do note that we only touched on the basics to get started. We only have an on.create handler, and no on.update handler. So if you want to increase the number of replicas, you will have to delete the custom resource and create a new one. Based on the example though, it should be pretty easy to fix that. The git repo contains an example of an operator that also implements the on.update handler (

%d bloggers like this: