Build Slack AI agents that post, answer and trigger workflows

What you can build

How to connect Slack

1. 01Create a Slack app with a bot token scoped to what the agent needs, for example chat:write and channels:history. Keep the token out of prompts; it lives on the tool's auth settings.
2. 02In AgentRow, create a tool of type HTTP API named slack_post_message with a POST to https://slack.com/api/chat.postMessage and an input schema for channel and text.
3. 03Set the auth type to bearer token and paste the bot token. Restrict allowed domains to slack.com.
4. 04Choose the safety level write_safe and, if humans should confirm posts, mark the tool as requiring approval.
5. 05Test the tool from the toolbox with a channel id, then attach it to the agent in the Tools tab and try it in the test chat.

Example tool definition

The shape a tool takes in the AgentRow builder API. Credentials are stored on the tool's auth settings, not in its configuration, so they never reach a prompt. Placeholders in double braces are filled by the model from the input schema.

{
  "name": "slack_post_message",
  "type": "http_api",
  "description": "Post a message to a Slack channel. Use after the user confirms the text and the channel.",
  "auth_type": "bearer_token",
  "safety_level": "write_safe",
  "requires_approval": false,
  "config": {
    "method": "POST",
    "url": "https://slack.com/api/chat.postMessage",
    "headers": { "Content-Type": "application/json" },
    "body_template": { "channel": "{{channel}}", "text": "{{text}}" },
    "allowed_domains": ["slack.com"]
  },
  "input_schema": {
    "type": "object",
    "properties": {
      "channel": { "type": "string", "description": "Channel id, for example C0123456789" },
      "text": { "type": "string", "description": "Message text in Slack mrkdwn" }
    },
    "required": ["channel", "text"]
  }
}

What the run trace shows

Every call appears in the run trace with the rendered request, the response status and body, timing, retries and errors, plus what the model was shown of the result. If the model never called the tool, the toolbox step shows whether it was bound for that model call and why.