This example is a chatbot for slack that interfaces with am LLM (GPT-4o) and maintains various tasks, like reminders or watching flight prices. The video below (click the image) has a nice intro to the functionality and architecture.
The core chatbot logic is in chat.ts. This uses Restate's Virtual Objects to have a chat session per user, with history and active tasks easily maintained in the object state.
The workflow-as-code execution of handlers ensures strict consistency and the single-writer semantics eforce a consistent message history even under concurrent messages or task notifications.
The tasks are simple Restate workflows. See the flight watch task as an example.
Finally, the slack integration handles slack webhooks, API calls, verification, deduplication, tracking message timestamps, etc.
You need an OpenAI access key to access GPT-4o. Export it as an environment variable like export OPENAI_API_KEY=sk-proj-...
Start the Restate Server (see "Get Restate" if you don't have it downloaded, yet.)
Make sure you install the dependencies via npm install.
Start the reminder and flight-watch tasks via npm run reminder-task and npm run flights-task respectively
(e.g., in different shells). In this example, we run them as separate endpoints. Since they are mostly suspending,
they would be prime candidates to be deployed on a FaaS platform.
Let Restate know about your task services via
restate dep add -y localhost:9081
restate dep add -y localhost:9082Start the main chatbot service via npm run app and register it at Restate Server restate dep add -y localhost:9080.
To chat with the bot, make calls to the chatbot service, like
curl localhost:8080/chatSession/<session-name>/chatMessage --json '"Hey, I am Malik, what tasks can you do?"'
curl localhost:8080/chatSession/<session-name>/chatMessage --json '"Create a reminder for in 30 minutes to get a coffee."'
The <session-name> path identifies the session - each one separately maintains its tasks and history.
Async notifications from tasks (like that a cheap flight was found) come in the chat bot's log, which is a bit hidden, but a result of the fact that this was initially written to be used with a chat app like Slack.
The chat bot can also be used as a Slack bot, as shown in the video. Each slack channel that the bot participates in and each direct message user are separate chat sessions.
The setup is a bit more involved, because you need to create a Slack App as the connection between Slack and the bot. This tutorial is a starting point.
For those with some experience in building slack apps, the requirements for this bot are:
- The following Bot Token OAuth Scopes:
channels:history,chat:write,groups:history,im:history,im:read,im:write,mpim:history - Event subscription for the following Bot Events:
message.channels,message.groups,message.im,message.mpim.
After installing the app to your workspace, export the following tokens and IDs as environment variables:
export SLACK_BOT_USER_ID=U...
export SLACK_BOT_TOKEN=xoxb-...
export SLACK_SIGNING_SECRET=...Once all keys are set up, start the app together with the slack adapter: npm run app -- SLACK.
Use a publicly reachable Restate server URL as Slack's event Request URL: https://my-restate-uri:8080/slackbot/message