StyloDocs
Recipes

Smart Auto-Reply with Knowledge Base

Search your knowledge base for relevant articles, generate a grounded AI draft reply, and surface it in the Zendesk agent sidebar for one-click approval.

View as Markdown

Search your knowledge base for relevant articles, generate a grounded AI draft reply, and surface it in the Zendesk agent sidebar for one-click approval.

Prerequisites: Connect Zendesk before adding this recipe. See Integrations.

How it works

Each box is a block; arrows show the order Stylo runs them, and branch labels show the path taken on each outcome.

Parameters

Set these when you add the recipe to your workspace:

ParameterTypeDefault
min_kb_confidencenumber0.4
brand_voicestringFriendly, professional, and empathetic. Use the customer's first name when available. Keep responses concise but thorough.

Blocks used

Every block in this recipe, linked to its reference page where one exists:

BlockTypeWhat it does
Webhooktrigger/webhookTrigger workflow via HTTP webhook
Get Ticket Contextzendesk/get-ticket-contextGet complete ticket context (ticket, conversation, customer, assignee, organization) optimized for AI
Search Knowledge Basecontext-engine/search-knowledge-baseSearch synced knowledge base documents using semantic search with automatic reranking
Conditionsystem/conditionBranch the workflow based on a boolean expression
Generate Textai/generate-textGenerate text using AI models
Human Reviewsystem/human-reviewPause and wait for human approval before continuing
Add Comment / Replyzendesk/add-commentAdd a public reply or internal note to a ticket

When to use it

Use this when you want agents to start from a knowledge-grounded draft instead of a blank reply box. It searches your knowledge base, drafts a reply only when a confident match exists, and surfaces it in the Zendesk sidebar for an agent to review.

Adapt it

This recipe is a starting point. To fit it to your workspace:

  • Tune the min_kb_confidence parameter to control how strong a knowledge-base match must be before a draft is generated.
  • Edit the brand_voice parameter to set the tone and style of the generated draft.
  • Adjust the confidence condition branch if you want different behavior when no confident match is found.

On this page