Get Answers in Slack 24/7 with this Custom Agent

A custom “customer support bot” built as a Notion custom agent now handles Slack questions automatically—answering from the company’s documentation, citing the exact source, and routing unanswered questions into a documentation update workflow. The practical payoff is speed and consistency: support teams can get answers in seconds for common, hard-to-find issues, while every unanswered question becomes a structured prompt to improve the knowledge base.

In Slack, the bot sits in a dedicated channel (e.g., “customer support ask”) and responds when someone posts a question. For a documented product query—such as how to store prior versions of a page—the bot searches the knowledge base, posts an answer in the thread, includes a citation (in this case pointing to the relevant “page settings” content), and marks resolution with a thumbs up/down. When the question is already covered, the bot’s response can be verified directly by opening the cited page, where the underlying “version history” information is visible.

When a question has no matching documentation, the bot doesn’t guess. Instead, it reports that it couldn’t find anything in the help center and creates a new knowledge base update request. That request becomes a ticket in a database (example: “Next big user conference date”), complete with suggested title and suggested content, plus a concrete implementation plan for how the documentation should be updated. A green check mark indicates the thread is resolved even when the outcome is “create a documentation update,” turning gaps in coverage into actionable work.

Behind the scenes, the agent is configured inside a Notion workspace under an “agents” sidebar area. The setup is organized into four main components: triggers (what events start the agent, such as a message in a Slack channel or a reaction), instructions (the agent’s operational “brain,” including what to do when answers are found vs. not found), tools and access (what the agent can view, edit, or create), and the model (the AI model used, with the option to choose from available leaders and update as offerings change).

The instructions are intentionally prescriptive. At a high level, they define the bot’s responsibility, the resources it should search (specific documentation pages), the exact behavior for answer-found versus answer-not-found paths, and how to handle Slack reactions and thread context. The agent also follows style and tone rules for how responses should read. Notably, creating a new custom agent can bootstrap its own prompt structure—so the initial configuration is generated in a familiar format rather than requiring a fully manual build.

Finally, the approach is presented as reusable across teams. Any group that fields questions in Slack—people operations, sales, product management—can build similar agents that search internal docs, cite sources, and open structured update requests when information is missing. The core idea is simple: automate retrieval and triage in Slack, then continuously improve the documentation based on real questions from the business.