Chat assistant

Log into Writer. From the Home screen, click on the Build an app button.

The Start building menu will appear, presenting options for the types of apps you can create.

Select Framework, located under Developer tools. This will create a brand new app based on Writer Framework.

On the next screen, titled How to deploy an application, you can get the API key for the app by clicking on the Reveal key button, located under the text Authenticate with an API key. Your complete API key will be displayed, and a “copy” button will appear. Click this button to copy the key; you’ll use it in the next step.

Open your terminal application. On macOS and Linux, this application goes by the name Terminal; on Windows, you can use either Windows PowerShell (which is preferred) or Command Prompt.

Install the writer and python-dotenv packages by entering the following commands in your terminal application:

pip install writer python-dotenv

This command tells pip, the Python package installer, to install two packages:

• writer, which provides some command-line commands and enables Python code to interact with Writer and the Writer Framework.
• python-dotenv, which makes it easy to manage environment variables by loading them from a .env file. This one is optional for this exercise, but you might find it useful when working on larger projects.

To pass your API key to the Writer Framework, you need to set an environment variable called WRITER_API_KEY.

Select your operating system and terminal application below, then copy and paste the command into your terminal application, replacing [your_api_key] with the API key you copied earlier:

The WRITER_API_KEY environment variable will remain defined as long your terminal session is open (that is, until you close your terminal application’s window).

Create the project by entering this command into your terminal application:

writer create chat-assistant --template=ai-starter

This command sets up a new project called chat-assistant using a starter template called ai-starter so that you’re not starting “from scratch.”

Select the Header component by clicking it — it’s the component at the top, containing the title AI STARTER and a gray area labeled Empty Header.

When you click it, you’ll see the properties panel appear on the right side of the page. This lets you view and edit the properties of the selected component.

The first property you’ll see in the panel is the Text property, which defines the text that appears as the header’s title. It should contain the value @{my_app.title}. The @{ and } indicate that my_app.title is a variable and that its contents should be the text displayed instead of the literal text “my_app.title”. You’ll set the value of this variable soon.

Select the Section component by clicking it — it’s just below the Header component and contains the title Section Title and a gray area labeled Empty Section.

In the properties panel, clear out the value of the Title property. This will remove the Section’s default title.

Drag a Text component from the Core toolkit panel on the left (it’s under Content, and you may need to scroll down a little to find it) and into the Section. Sections can act as containers for other components.

Select the Text component. In the properties panel, set the Text property to provide instructions or context for your chat assistant. Here’s an example: Welcome to the Chat Assistant. Ask me anything!

The heart of this app is the Chatbot component, a pre-built component that displays the conversation between the LLM and the user and provides a text field where the user can enter prompts.

Drag a Chatbot component from the Core toolkit panel (it’s under Content) into the Section, just below the Text box.

You should see the following at the start of the file:

import writer as wf
import writer.ai

Replace that code with the following:

import os
import writer as wf
import writer.ai

# Set the API key
wf.api_key = os.getenv("WRITER_API_KEY")

This code imports the libraries that the application will need and then reads your Writer Framework API key in the WRITER_API_KEY environment variable.

The application needs a function to handle incoming chat messages. Find these comments in the code…

# Welcome to Writer Framework! 
# This template is a starting point for your AI apps.
# More documentation is available at https://dev.writer.com/framework

…and replace them with the following function:

def generate_completion(state, payload):
  print(f"Here's what the user entered: {payload['content']}")
  state["conversation"] += payload
  print(f"Conversation: {state['conversation'].messages}")
  try:
      for index, chunk in enumerate(state["conversation"].stream_complete()):
          print(f"Chunk {index}: {chunk}")
          if not chunk.get("content"):
              chunk["content"] = ""
          state["conversation"] += chunk
          
      print(f"state['conversation']:\n{state['conversation'].messages}")
  except Exception as e:
      print(f"Error during stream_complete: {e}")

The generate_completion() function will be called when the user enters a prompt, which is contained in the payload object. The payload object is added to the conversation object contained in the application’s state, which adds the user’s prompt to the record of the conversation between the user and the LLM.

After adding the user’s prompt to the conversational record, generate_completion() calls the conversation object’s stream_complete() method, which generates an LLM completion based on the conversation so far. As its name implies, stream_complete() returns the completion as a stream of text chunks, which are captured and added to the conversation object.

Note that generate_completion() completion uses a lot of print() functions for debugging purposes, and you can use them to get a better idea of what’s happening in the function. You’ll see their output in both your terminal application and in the project editor’s ‘log’ pane (which will be covered shortly) as you use the chat assistant. This output will include:

• The prompt the user entered
• The chunks of data that make up the LLM’s response as they are generated
• The record of the conversation between the user and the LLM.

The print() functions don’t affect the operation of the chat assistant in any way, and you can remove them if you wish.

The final step is to set the application’s initial state. Find this code, which should be just after the generate_completion() function…

# Initialise the state
wf.init_state({
    "my_app": {
        "title": "AI STARTER"
    },
})

…and replace it with this:

# Initialize the state
wf.init_state({
    "conversation": writer.ai.Conversation(),
    "my_app": {
        "title": "CHAT ASSISTANT"
    },
})

The Writer Framework’s init_state() method sets the initial value of state, a dictionary containing values that define the state of the application. The key-value pairs in state are how you store values used by your app and how you pass data between the back-end code and the UI.

The code above sets the initial value of state so that it has two key-value pairs:

• conversation: An object that keeps a record of the conversation that the user is having with the LLM. You’ll bind its value to the Chatbot component soon.
• my_app: A dictionary containing values that define the application’s appearance. This dictionary has a single key-value pair, title, which defines the text that appears as the application’s title in the Header.

That’s all the code. If you edited the code in the browser, save it by clicking the “save” button near the top right corner of the code editor.

Click the “toggle code” button to hide the code editor.

Earlier, you saw that the Header component’s Text property was set to @{my_app.title}, a value in the app’s state variable. You changed this value when you update the call to the Writer Framework’s init_state() method.

Recall that the conversation object contained within the state variable contains the record of the conversation that the user is having with the LLM. Binding the Chatbot component to this object allows it to display the conversation to the user.

Select the Chatbot component. In the properties panel, find the Conversation property and set its value to @{conversation}.

The value @{conversation} specifies that the Chatbot component should get its information from the value corresponding to the conversation key in the application’s state variable.

You need to specify that the Chatbot component should call the generate_completion() function when the user enters a prompt.

Do this by scrolling down the properties panel to the Events section until you see a property called wf_chatbot_message. Select generate_completion from its menu.