Skip to content
GitHub

Chatbot settings

How the Chatbots tab organizes an agent's deployments, and what each connection's Connection, Trigger, Preferences and Activities tabs control.


On this page

The Chatbots tab is where an agent is deployed. Each deployment is a connection: one agent in one place, such as a Slack channel, a direct-message bot, a Microsoft Teams team or chat, a Discord server, a HubSpot account or a Zendesk account. An agent can have many connections, and each one keeps its own triggers and preferences, so the same agent can answer every message in a support channel while waiting for an @mention everywhere else.

This page covers what you can configure on a connection. For the install steps of each channel, see Slack, Microsoft Teams, Discord, HubSpot and Zendesk.

Adding a connection#

Open the agent's Chatbots tab and click a channel's tile. Which tiles you can use depends on two things:

  • The agent's runtime. Agents on the Claude Agent SDK runtime deploy to Slack and Microsoft Teams only, so the other tiles aren't shown. See Runtimes.
  • Your plan. A channel your plan doesn't include shows a lock. Clicking it opens the plan picker instead of the setup flow. See channels by plan.

Once an agent has connections, the selector at the top of the tab switches between them, and its Connect New Channel option adds another.

Connection tabs#

Selecting a connection opens its tabs:

TabWhat it holdsWhere it appears
ConnectionWhere the agent is deployed, such as which Slack channels or which Teams team or chatEvery channel except Zendesk
TriggerWhat makes the agent respondEvery channel
PreferencesHow replies look and what context the agent receivesSlack, Microsoft Teams and Zendesk, except on Claude Agent SDK agents
ActivitiesThe runs this connection started, with the trace behind each oneEvery agent except OpenAI Responses agents

The Preferences tab is hidden for agents on the Claude Agent SDK runtime, because that runtime doesn't read these options yet.

Restricting who can use a Slack connection#

Where it is enabled, a Slack connection's Connection tab has a Restrict to specific users switch ("Control which Slack users can interact with this bot"). While it is off, everyone in the workspace can use the agent in the channels it has access to. Turn it on and pick users, and only those users' messages, mentions, reactions and slash commands start the agent on that connection; the same events from anyone else are ignored.

The agent's Access control setting governs the dashboard only; it doesn't change who can talk to the agent in Slack.

Triggers#

Slack triggers#

TriggerWhat it does
Always replies to @mentions and DMsAlways on, with no switch. The agent replies whenever someone @mentions it in a channel or sends it a direct message
Continue in threads (after first reply)Once the agent has joined a thread, it keeps responding there without another @mention. Choose AI-powered smart replies (Experimental), which answers only when it has something useful to add, or Respond to all replies (Deprecated)
Trigger on a scheduleThe agent posts on its own at a time or interval you choose, with the prompt you give it. Not offered on a direct-message connection or one that covers all channels
Reply without @mentionThe agent also replies to every new message in the channel. High traffic, so use it sparingly
Trigger by keywordThe agent replies when a message contains one of your keywords
Trigger by emoji reactionThe agent replies when someone reacts to a message with one of your chosen emoji
Trigger by senderThe agent replies whenever a specific user or bot posts, for example to follow up on an alerting bot

If the Slack installation lacks the permissions the emoji or sender trigger needs, turning it on asks you to reinstall the app from Settings → Integrations → Slack first. Only the person who installed it can reinstall it.

Agents on the Claude Agent SDK runtime offer only Always replies to @mentions and DMs, Continue in threads (after first reply), which uses AI-powered smart replies, Reply without @mention and Trigger by keyword.

Microsoft Teams triggers#

  • Always replies to @mentions and DMs (always on)
  • Continue in threads (every reply): once a thread has started, the agent keeps responding to new messages in it
  • Reply without @mention: the agent also replies to every new message in the channel
  • Trigger by keyword

Discord triggers#

  • Always replies to @mentions and DMs (always on)
  • Continue in threads (every reply)

HubSpot trigger#

The agent runs when a ticket is created and writes its answer to a Runbear card on the ticket. This trigger is always on. See HubSpot.

Zendesk triggers#

Enable one or both:

  • Trigger by ticket creation. The agent responds when a new ticket is created. Narrow it with Tag filters (only tickets with those tags) and Agent filters (only tickets assigned to those Zendesk agents), and optionally give it a Custom prompt of up to 1,024 characters that replaces the default instruction for this trigger.
  • Trigger by internal comment. The agent responds to an internal comment that starts with the Trigger word, which defaults to @Runbear. For example, an internal note starting with @Runbear summarize this ticket.

Preferences#

Slack preferences#

PreferenceWhat it does
App iconThe bot's profile image. For the shared Runbear bot, upload an image here. For a custom bot, the row links to the bot's settings on api.slack.com, where the icon is managed
Use full thread as contextOn: the agent reads every message in the thread. Off: only the message that triggered it
Only visible to selected usersReplies as a Slack ephemeral message. Pick the Audience; leave it empty to show the reply only to the person who triggered the agent
Hide Runbear links in error messagesStrips links to Runbear from error messages
Custom error messageReplaces the agent's default error message with your text
Hide source referencesOmits citation markers and the source list from replies
Send username with each requestPasses the Slack user's name to the agent so it can personalize replies
Custom message templateWraps each user message before it reaches the agent, using variables such as {message} and {user_id}
Reply as new channel messagePosts replies as new channel messages instead of in a thread. A checkbox extends this to messages that triggered the agent inside a thread
Welcome new users on first interactionSends a greeting the first time a Slack user triggers the agent

Microsoft Teams preferences#

Teams offers Use full thread as context, Hide Runbear links in error messages, Hide source references, Send username with each request, Custom message template and Reply as new channel message, which work as in Slack, plus Hide agent name in replies, which omits the agent's name from the responses it posts.

Zendesk preferences#

  • Reply as public ticket comment. By default the agent posts private internal notes that only your support team sees. Turn this on to post customer-facing replies instead. The connection form calls it Reply as public comment.
  • Hide source references. Omits citation markers and the source list.
  • Reply only when knowledge base has an answer. The agent stays silent unless it finds supporting information in the agent's knowledge sources. The connection form calls it Reply only with knowledge base.

Deleting a connection#

In the Chatbots tab, select the connection, click the trash icon (Delete the selected connection), and confirm with Delete. The agent stops responding there, and the connection's configuration is deleted. The agent itself and its other connections are unchanged.

To remove a bot or workspace from your whole organization instead, see Integrations.