AI Chatbot

In this tutorial, you'll learn how to bring the power of AI into your Slack workspace using a chatbot called Bolty that uses Anthropic or OpenAI.

With Bolty, users can:

send direct messages to Bolty and get AI-powered responses in response,
use the /ask-bolty slash command to ask Bolty questions, and
receive channel summaries when joining new channels.

Intrigued? First, grab your tools by following the three steps below.

import QuickstartGuide from '@site/src/components/QuickstartGuide';

Prerequisites {#prereqs}

You will also need the following:

a development workspace where you have permissions to install apps. If you don’t have a workspace you can join the Developer Program and provision a sandbox with access to all Slack features for free.
a development environment with Python 3.7 or later.
an Anthropic or OpenAI account with sufficient credits, and in which you have generated a secret key.

Obtaining and storing your environment variables {#environment-variables}

Before you'll be able to successfully run the app, you'll need to first obtain and set some environment variables.

Provider tokens {#provider-tokens}

Models from different AI providers are available if the corresponding environment variable is added as shown in the sections below.

To interact with Anthropic models, navigate to your Anthropic account dashboard to create an API key, then export the key as follows:

bash

export ANTHROPIC_API_KEY=<your-api-key>

</TabItem> <TabItem value="Google Cloud Vertex AI" label="Google Cloud Vertex AI">

To use Google Cloud Vertex AI, follow this quick start to create a project for sending requests to the Gemini API, then gather Application Default Credentials with the strategy to match your development environment.

Once your project and credentials are configured, export environment variables to select from Gemini models:

bash

export VERTEX_AI_PROJECT_ID=<your-project-id>
export VERTEX_AI_LOCATION=<location-to-deploy-model>

The project location can be located under the Region on the Vertex AI dashboard, as well as more details about available Gemini models.

</TabItem> <TabItem value="OpenAI" label="OpenAI">

Unlock the OpenAI models from your OpenAI account dashboard by clicking create a new secret key, then export the key like so:

bash

export OPENAI_API_KEY=<your-api-key>

</TabItem> </Tabs>

Setting up and running your local project {#configure-project}

Start your Python virtual environment:

bash

python3 -m venv .venv
source .venv/bin/activate

</TabItem> <TabItem value="windows" label="Windows">

bash

py -m venv .venv
.venv\Scripts\activate

</TabItem> </Tabs>

Install the required dependencies:

bash

pip install -r requirements.txt

Run your app locally:

bash

slack run

If your app is indeed up and running, you'll see a message that says "⚡️ Bolt app is running!"

Choosing your provider {#provider}

Navigate to the Bolty App Home and select a provider from the drop-down menu. The options listed will be dependent on which secret keys you added when setting your environment variables.

If you don't see Bolty listed under Apps in your workspace right away, never fear! You can mention @Bolty in a public channel to add the app, then navigate to your App Home.

Setting up your workflow {#workflow}

Within your development workspace, open Workflow Builder by clicking on your workspace name and then Tools > Workflow Builder. Select New Workflow > Build Workflow.

Click Untitled Workflow at the top to rename your workflow. For this tutorial, we'll call the workflow Welcome to the channel. Enter a description, such as Summarizes channels for new members, and click Save.

Select Choose an event under Start the workflow..., and then choose When a person joins a channel. Select the channel name from the drop-down menu and click Save.

Under Then, do these things, click Add steps and complete the following:

Select Messages > Send a message to a person.
Under Select a member, choose The user who joined the channel from the drop-down menu.
Under Add a message, enter a short message, such as Hi! Welcome to {}The channel that the user joined. Would you like a summary of the recent conversation? Note that the {}The channel that the user joined is a variable; you can insert it by selecting {}Insert a variable at the bottom of the message text box.
Select the Add Button button, and name the button Yes, give me a summary. Click Done.

We'll add two more steps under the Then, do these things section.

First, scroll to the bottom of the list of steps and choose Custom, then choose Bolty and Bolty Custom Function. In the Channel drop-down menu, select Channel that the user joined. Click Save.

For the final step, complete the following:

Choose Messages and then Send a message to a person. Under Select a member, choose Person who clicked the button from the drop-down menu.
Under Add a message, click Insert a variable and choose {}Summary under the Bolty Custom Function section in the list that appears. Click Save.

When finished, click Finish Up, then click Publish to make the workflow available in your workspace.

Interacting with Bolty {#interact}

Summarizing recent conversations {#summarize}

In order for Bolty to provide summaries of recent conversation in a channel, Bolty must be a member of that channel.

Invite Bolty to a channel that you are able to leave and rejoin (for example, not the #general channel or a private channel someone else created) by mentioning the app in the channel — i.e., tagging @Bolty in the channel and sending your message.
Slackbot will prompt you to either invite Bolty to the channel, or do nothing. Click Invite Them. Now when new users join the channel, the workflow you just created will be kicked off.

To test this, leave the channel you just invited Bolty to and rejoin it. This will kick off your workflow and you'll receive a direct message from Welcome to the channel. Click the Yes, give me a summary button, and Bolty will summarize the recent conversations in the channel you joined.

The central part of this functionality is shown in the following code snippet. Note the use of the user_context object, a Slack type that represents the user who is interacting with our workflow, as well as the history of the channel that will be summarized, which includes the ten most recent messages.

python

from ai.providers import get_provider_response
from logging import Logger
from slack_bolt import Complete, Fail, Ack
from slack_sdk import WebClient
from ..listener_utils.listener_constants import SUMMARIZE_CHANNEL_WORKFLOW
from ..listener_utils.parse_conversation import parse_conversation

"""
Handles the event to summarize a Slack channel's conversation history.
It retrieves the conversation history, parses it, generates a summary using an AI response,
and completes the workflow with the summary or fails if an error occurs.
"""

def handle_summary_function_callback(
    ack: Ack, inputs: dict, fail: Fail, logger: Logger, client: WebClient, complete: Complete
):
    ack()
    try:
        user_context = inputs["user_context"]
        channel_id = inputs["channel_id"]
        history = client.conversations_history(channel=channel_id, limit=10)["messages"]
        conversation = parse_conversation(history)

        summary = get_provider_response(user_context["id"], SUMMARIZE_CHANNEL_WORKFLOW, conversation)

        complete({"user_context": user_context, "response": summary})
    except Exception as e:
        logger.exception(e)
        fail(e)

Asking Bolty a question {#ask-app}

To ask Bolty a question, you can chat with Bolty in any channel the app is in. Use the \ask-bolty slash command to provide a prompt for Bolty to answer. Note that Bolty is currently not supported in threads.

You can also navigate to Bolty in your Apps list and select the Messages tab to chat with Bolty directly.

Next steps {#next-steps}

Congratulations! You've successfully integrated the power of AI into your workspace. Check out these links to take the next steps in your Bolt for Python journey.

To learn more about Bolt for Python, refer to the Getting started documentation.
For more details about creating workflow steps using the Bolt SDK, refer to the workflow steps for Bolt guide.