Skip to Content
ExamplesDeflect L1 from your docs

📚 Deflect L1 questions from your docs

StackTrigger: Slack / Discord · Models: your choice · Tools: docs search, ticket-create API · Mode: single agent + constraints · Memory: shared · Output: answer in-channel, ticket only if it can't

Who it’s for — teams whose docs already hold the answer, but people ask in Slack anyway. What it does — answers product and setup questions from your own documentation in the channel, cites the page it used, and only opens a ticket when it genuinely can’t answer — so humans see the hard questions, not the FAQ.

Build it

  1. Make your docs searchable. Expose your help center or docs as a search tool — connect a search API (Algolia, your own endpoint) via Connect API, or wrap a retrieval endpoint as a JavaScript tool.
  2. Create the agent with the instruction: answer only from the docs; always cite the page; if the docs don’t cover it, say so and open a ticket.
  3. Add a constraintnever invent a feature or a limit that isn’t in the docs — so it can’t hallucinate its way to a confident wrong answer.
  4. Give it a fallback tool — your ticket-create API, so an unanswerable question becomes a real ticket with the conversation attached.
  5. Turn on shared memory so recurring gaps (“everyone asks about SSO setup”) accumulate and point you at the docs worth writing.
  6. Publish it on Slack or Discord. Mention the bot, it answers inline with a citation.

Make it yours

  • Point it at internal runbooks instead of public docs for an ops helpdesk.
  • Add a verify step so a second model checks the citation actually supports the answer.
  • Log every “docs didn’t cover it” to a channel as a content backlog.

UsesSupport · Slack · Discord · Constraints · Memory