Front and Knowledge: AI Suggested-Reply Generator Template

When a new conversation lands in Front, this Spojit template drafts a grounded reply from your help collection and adds it to the conversation for a teammate to review.

What It Builds

A Webhook trigger fires when Front reports a new conversation, a Knowledge node in Query mode drafts a suggested reply grounded in your help collection, and a Connector node on the front connector adds that draft to the same conversation. The draft is never sent automatically: it lands in Front so a teammate can read, edit, and send it, which keeps a human in the loop while removing the blank-page problem on every new message.

The Prompt

Paste this into Miraxa, the intelligent layer across your automation, and it builds the workflow, connecting the tools for you:

Build a workflow that triggers on a Front webhook when a new conversation is created. Take the customer's message and use a Knowledge node in Query mode against my "support-docs" collection to draft a grounded, friendly suggested reply with citations. Then use the Front connector to add that draft as a comment on the same conversation so a teammate can review and send it.

Connectors Used

  • Webhook trigger - starts the run when Front posts a new-conversation event; the conversation id and customer message arrive as {{ input }}.
  • Knowledge node (Query mode) - searches your persistent help collection and drafts the reply with a Prompt, a Result Count, and a synthesis Model.
  • front - the get-conversation tool pulls the latest message, and send-message adds the drafted reply back to the conversation for review.

Customize It

Swap support-docs for the name of your own Knowledge collection, and tune the reply by editing the Prompt and Result Count in the Knowledge node's properties panel (raise the count for broader context, lower it for tighter answers). To draft in a specific tone or language, add those instructions to the Knowledge Prompt. If you would rather post the draft as a private internal note than a customer-visible message, adjust how the front node maps its inputs.

Tips

  • Keep the front step in Direct mode so the reply always lands on the right conversation: it is a single, predictable call to send-message with no AI cost.
  • Embed your help articles into the collection first (the Knowledge node can only ground answers in documents you have already uploaded), and query with the same embedding model you embedded with.
  • Use the Knowledge node's Response Schema if you want the draft returned as structured JSON (for example a reply field plus a list of source titles) before you format it for Front.

Common Pitfalls

  • The Webhook trigger needs a signing connection to verify Front's requests; pick or create one when you set up the trigger, or the run will reject the incoming event.
  • An empty or stale collection produces vague drafts. Refresh the collection whenever your help content changes so suggestions stay accurate.
  • Because the draft posts back to the same conversation, test on a low-volume inbox first so reviewers are not surprised by suggested replies appearing on live customer threads.

Related Articles

Did this answer your question? Thanks for the feedback There was a problem submitting your feedback. Please try again later.