The Slack Bot

The Smarter Helm chart can run an optional Slack bot, which answers questions in a Slack workspace with a deployed LLMClient. The Smarter Project uses it to bring smarter_developer, the LLMClient that answers from the Smarter Project MCP Server, into its own Slack workspace, but any deployed LLMClient that doesn’t require an api key works.

People talk to the bot in two ways:

  • @mention it in a channel. It answers in the message’s thread, and every later mention in that thread continues the same conversation.

  • Send it a direct message. The messages outside threads are one conversation, and each thread is a conversation of its own.

A message that is only new, reset, start over or new conversation starts a new conversation in its thread. Smarter on Slack is the guide for the workspace’s members, to which new members can be referred.

How It Works

The bot is the chart’s slack-bot subchart (helm/charts/slack-bot), listed in the smarter chart’s Chart.yaml with the condition slackBot.enabled, which is false by default. When it is enabled, the chart adds one Deployment, smarter-slack-bot, and a ConfigMap that holds the bot’s program, files/bot.py.

  • Socket Mode. The bot opens its own connection to Slack, so nothing connects to it: there is no Service, Ingress, DNS record or certificate. It needs only outbound HTTPS, to Slack and to the LLMClient.

  • The Smarter Chat api. For each new Slack thread, the bot starts a chat session with POST <agentUrl>/config/, and asks each question with POST <agentUrl>/prompt/, sending the thread’s questions and answers. The platform runs the LLMClient’s whole prompt, tools and Guardrails included, and records it in the session, so every conversation appears in the LLMClient’s chat history and cost accounting like any other. See The Chat API Contract.

  • One replica. The bot keeps each thread’s chat session in memory, for its 500 most recently used threads, so a restart starts every conversation again. Two pods would each receive part of a thread’s messages, so the Deployment runs one replica and is replaced with Recreate.

  • A stock Python image. The bot needs only the standard library and Slack Bolt, which the container installs when it starts (slackBot.slackBoltVersion). It runs as a non-root user, with a read-only root file system and no Kubernetes service account token, and receives only the two Slack tokens.

Every Slack message the bot answers is a prompt, billed by the LLMClient’s provider, and anyone in the workspace can send one. Choose an LLMClient whose model and budget suit that.

Setting It Up

  1. Create the Slack app at api.slack.com/apps, from scratch, in your workspace.

    • Under Socket Mode, enable Socket Mode and create an app-level token with the connections:write scope. This is the app token, xapp-....

    • Under OAuth & Permissions, add the bot token scopes app_mentions:read, chat:write and im:history.

    • Under Event Subscriptions, subscribe to the bot events app_mention and message.im.

    • Under App Home, enable the Messages tab, so that people can send the bot direct messages.

    • Install the app to the workspace. This gives the bot token, xoxb-....

  2. Choose the LLMClient. Deploy it, and check that its own host answers without an api key:

    curl -X POST -H "Content-Type: application/json" -d '{}' \
  https://smarter-developer.<account-number>.api.<root-domain>/config/

  3. Configure the environment that runs the bot, and only that one: two bots with the same Slack app would each receive part of the messages. In the repository’s GitHub settings, under the environment (e.g. prod), add:

    • the secrets SLACK_BOT_TOKEN and SLACK_APP_TOKEN;

    • the variable SLACK_BOT_AGENT_URL, the LLMClient’s url.

    The deploy workflow passes the tokens to the chart as env.SMARTER_SLACK_BOT_TOKEN and env.SMARTER_SLACK_APP_TOKEN, which are stored in the platform’s environment Secret, and the url as slackBot.agentUrl. It sets slackBot.enabled to true only when all three are set, so the other environments don’t run the bot.

  4. Deploy, and invite the bot to a channel (/invite @<bot name>).

To install the chart by hand instead, set the same values:

helm upgrade --install smarter ./helm/charts/smarter/ \
  --set slackBot.enabled=true \
  --set slackBot.agentUrl=https://smarter-developer.<account-number>.api.<root-domain>/ \
  --set env.SMARTER_SLACK_BOT_TOKEN=xoxb-... \
  --set env.SMARTER_SLACK_APP_TOKEN=xapp-... \
  ...

Values

Value

Meaning

slackBot.enabled

false by default. true deploys the bot.

slackBot.agentUrl

Required when enabled. The deployed LLMClient’s url.

slackBot.timeout

Seconds to wait for an answer, 240 by default. An answer with several tool calls takes 10 to 20 seconds.

slackBot.slackBoltVersion

The Slack Bolt release that the container installs.

slackBot.image

The Python image, python:3.12-slim by default.

slackBot.resources

The container’s requests and limits: 50m CPU and 128Mi of memory, up to 256Mi.

env.SMARTER_SLACK_BOT_TOKEN, env.SMARTER_SLACK_APP_TOKEN

The Slack app’s bot and app-level tokens.

Operating It

Read the bot’s log, which records every question it could not answer:

kubectl logs -n <namespace> deploy/smarter-slack-bot

When the LLMClient doesn’t answer, the bot apologizes in Slack and logs the error. When Slack’s connection drops, Slack Bolt reconnects by itself. A pod that is replaced, for example with its node, forgets its conversations: people continue by mentioning the bot again.

The bot’s tests (helm/charts/slack-bot/tests) need only the standard library, and run in the Tests workflow’s slack-bot job:

python -m unittest discover -s helm/charts/slack-bot/tests