Compose domain specific bots using an agent bot

A domain specific bot addresses queries related to a specific domain or topic. Some examples are - Travel bot(Travel related conversations), Weather bot (Weather related conversations).

If a user wants to have a cross-domain conversation, then the user will have to switch between the different bots.

There are scenarios where an user wants to have a conversation involving multiple domains. For example - When I want to travel to a place, I want to query for the weather and then book a cab or flight. I might have to end up switching between two bots, weather bot and travel bot. What if I could just have one interface bot which will redirect my messages to a specific bot and get answers to me?

Well, this code pattern showcases an implementation of this approach.

The solution here is to have an agent bot (or an interface bot) and a few other bots which can handle conversations for a specific domain - let's call these specific bots. The agent bot knows about the specific bots and also about which domain each of them can handle. When user initiates conversation with agent bot, the agent bot will understand the intent of user query and it will redirect the user query to a specific bot. Subsequent requests from user are redirected to specific bot. When the conversation with the specific bot is over or when the specific bot is not able to handle the request, the control is given back to agent bot which will then redirect the messages to appropriate bot.

This approach provides seamless experience for users. It can be used by organisations which provide a host of services to its customers like financial services, tours and travel agencies, news agencies etc..

Advantages with this approach are:

• Plug and play the bots
• Modular approach facilitates Bots composition
• Come up with new services by composing two or more Bots
• Easy to maintain, make changes, add/remove functionalities
• Easy to troubleshoot issues
• Transparent to user

In this code pattern we will use Watson assistant Bot for building Bots and Node.js application as orchestration layer.

When the reader has completed this Code Pattern, they will understand how to:

• How to configure a bot to make it Agent Bot
• How to configure a specific bot to return control to Agent Bot
• How to build an orchestration layer to stitch Agent Bot and specific Bots

Flow

1. User accesses web application and types in a message. Nodejs application, an orchestration later, sends user message to agent bot
2. Agent bot determines the intent of the message and responds with the specific bot details, to which the message needs to be redirected.
3. Node.js application sends message to the specific bot (Weather Bot, in this case). Specific bot responds. Conversation continues between user and specific bot.
4. When the conversation with specific bot is over, user message is then sent to agent bot to determine the intent.
5. Node.js application sends message to the specific bot (Travel Bot, in this case). Specific bot responds. Conversation continues between user and specific bot.

Watch the Video

Steps

1. Clone git repo.
2. Create bots.
3. Configure application with bots details.
4. Deploy application to IBM Cloud.
5. Run application.

1. Clone git repo

• On command prompt run the below command to clone the git repo.

git clone git@github.com:IBM/watson-assistant-multi-bot-agent.git

or

git clone https://github.com/IBM/watson-assistant-multi-bot-agent.git

run cd watson-assistant-multi-bot-agent to change directory to project parent folder

2. Create bots

2.1 Create Watson Assistant service instance

• Click this link to create Watson assistant service.
• Enter the service name as Watson Assistant-bots. You can choose to enter any name you like.
• Ensure you select the right region, organisation and space.
• Under Pricing Plans, select Lite plan.
• Click Create.
• Watson Asistant service instance should get created.

2.2 Import bots

• Go to IBM Cloud Resource list and click on the Watson Assistant service instance created in above steps.
• On the Watson Assistant Resource list page, click Launch Watson Assistant.

• Click Skills tab in the side bar.

• Click the Create skill button.
• Select the Dialog skill box
• Click the Next button.
• Select the Import skill tab.

• Click on Choose JSON file.
• Browse to the cloned repository parent folder -> WA.
• Select agent-bot.json and click Open.

• Click Import button.
• Repeat above steps in section Import bots to import travel_bot.json and weather_bot.json.

3. Configure application with bots details

3.1 Gather required details

• Go to IBM Cloud Resource list and click on the Watson Assistant service instance.
• On the Watson Assistant Resource list page, click Launch Watson Assistant.
• Click Skills tab in the side bar.

• On agentBot click actions, the three vertical dots on the top right corner.

• Click View API Details.

• Copy and save Skill ID for later use. It can be interchangeably used as Workspace ID.

• Repeat above steps in section Configure application with bots details for all the other bots also.

• Go to IBM Cloud Resource list and click on the Watson Assistant service instance.

• Copy API Key and Url using the copy buttons as shown in below image or using the Show Credentials and copying the field contents. Save them in a text file for later use.

3.2 Update manifest.yml file with the details gathered

• Under project parent folder, open manifest.yml file for editing.
• Update ASSISTANT_IAM_API_KEY with Watson Assistant service instance's API Key as noted in section Gather required details
• Update ASSISTANT_IAM_URL with Watson Assistant service instance's Url as noted in section Gather required details
• Update WORKSPACE_ID_AGENT, WORKSPACE_ID_TRAVEL, WORKSPACE_ID_WEATHER with Workspace IDs of respective bots as noted in section Gather required details

Updated manifest.yml file looks as below

4. Deploy application to IBM Cloud

• On command prompt, navigate to project parent folder
• On command prompt, login to IBM Cloud using ibmcloud login or ibmcloud login --sso (for federated login).
• Ensure that you are in the right organisation, space and region using the below command.

ibmcloud target

• Run the below command to deploy the application to IBM Cloud.

ibmcloud cf push

• Check the logs of the application using the command ibmcloud cf logs <app_name> --recent.
• Ensure that the application is deployed to IBM Cloud successfully. If you see any errors in logs, fix them and redeploy the application.

6. Run the application

• Login to IBM Cloud and from the dashboard.
• Click on Cloud Foundry apps. The application you deployed should be listed and it should be in running state.
• Click on the application and again click on Visit App URL.

• The application home page opens.

• On command prompt monitor logs using the command

ibmcloud cf logs watson-assistant-multi-bot-agent

• On the application home page type a weather related query What does the weather look like tomorrow?.
• Check the log files and notice that the message first goes to Agent Bot and then it is redirected to the Weather Bot.
• In the interface, you are asked to enter the location. Enter a location, e.g. Bengaluru.
• Check the logs. Because the conversation with the Weather Bot has not ended, subsequent messages are sent to the Weather Bot itself, without the intervention of the Agent Bot.
• The Weather Bot responds with an answer to user query and hence conversation with Weather Bot is treated as ended.
• Next user enters a travel related query Book a cab.
• In the logs, notice that the message is sent to Agent Bot and then it is redirected to Travel Bot.
• In the interface you are asked to enter a date for cab booking. Enter a date or you can just say Today.
• Check the logs. Because the conversation with the Travel Bot has not ended, subsequent messages are sent to the Weather Bot itself, without the intervention of the Agent Bot.
• In the interface you are asked to enter time for the cab to arrive. Say 12 PM.
• Check the logs. Travel Bot responds with an answer. The conversation with Travel Bot has ended.
• The above conversation flow can continue and the Agent Bot will redirect the messages to specific bots based on the intent of user query
• Some of the basic messages as greetings and bye can be handled by Agent Bot itself.
• In the interface type Thank you. Bye

• Overall Flow of messages between bots is as shown in the below diagram

Legend for above image

1. Message sent to Agent Bot.
2. Message redirected to Weather Bot.
3. Response from Weather Bot.
4. Message sent to Weather Bot.
5. Response from Weather Bot (end of conversation with Weather Bot).
6. Message sent to Agent Bot.
7. Message redirected to Travel Bot.
8. Response from Travel Bot.
9. Message sent to Travel Bot.
10. Response from Travel Bot.
11. Message sent to Travel Bot.
12. Response from Travel Bot (end of conversation with Travel Bot).
13. Message sent to Agent Bot.
14. Response from Agent Bot (end of conversation).

Plug and play process for a new bot

1. To add a new bot, create a new bot in the Watson Assistant service instance created earlier for this code pattern or you can use an existing bot that you want to use. Let's say you added a bot for Restaurant Booking and named it as RESTAURANT_BOOKING.
2. In manifest.yml file, add an entry for new bot as shown below

WORKSPACE_ID_RESTAURANT_BOOKING: xxxxxxxxxxxxxxxxxxxxxxxx

3. In the RESTAURANT_BOOKING Bot, when the conversation is over (last node in a dialog), add a context parameter destination_bot and value as AGENT. It enables the control to be passed to back to the Agent Bot. You can refer to the leaf nodes in other already imported bots for examples.
4. Open Agent Bot. Add an intent for Restaurant Booking, say restaurant. Then in dialog, add a node for restaurant intent. Add a context parameter destination_bot and value as RESTAURANT_BOOKING.
5. Redeploy the application for the configuration changes to take effect.
6. That's it, you are set to you new RESTAURANT_BOOKING bot as a plug and play bot.

Summary

We introduced an Agent Bot which understands intents of messages. Agent Bot will redirect a message to a specific bot which can handle that message. We saw how to configure Agent Bot, Specific Bots and orchestration application to have this arrangement possible. We also saw how to plug and play this setup to add a new bot.

Troubleshooting

• The application page displays but there is no greetings message, when the application pages loads.

  • Verify Watson Assistant username and password are correct.
  • Verify workspace_id of the agent bot is correct.

• No response for a chat message Verify workspace_id of a specific bots is correct.

• If there is any other issue, check application logs using the below command

ibmcloud cf logs <app-name> --recent

License

Apache 2.0