n8n template

Support Answers with Citations

A support assistant that shows where every answer came from, and hands the rest to a person instead of inventing one.

$29 one-time, USD

Secure checkout by Paddle, our merchant of record. Taxes are calculated at checkout. 14-day refund.

The workflow open on the n8n canvas
28 nodes with setup notes inside

Three outcomes. Never a guess.

Most support bots let the model decide whether it knows the answer. Here that decision is made by two checks written as plain code, so they behave the same every time.

A question is searched against your docs. A score gate in code either escalates to a person or lets the AI answer from the passages. A citation check in code either sends the cited answer or escalates. Customer askschat Search your docsscored matches Score gatecode, MIN_SCORE Answer frompassages only Citation checkcode Cited answersent Hand off to a personTelegram alert, and the customer is told a human will follow up docs silent ungrounded

The score gate

Every match comes back with a similarity score. Below your threshold, the AI is never called.

// ---- TUNE THESE TWO. ----
const MIN_SCORE = 0.7;   // how close a passage must be to count
const MAX_SOURCES = 5;   // most passages handed to the AI

const kept = scored
  .filter(m => m.score >= MIN_SCORE)
  .sort((a, b) => b.score - a.score)
  .slice(0, MAX_SOURCES);

const covered = kept.length > 0;

The citation check

The answer must cite [1], [2]. Every number must point to a passage that was really supplied.

const valid = new Set(ctx.sources.map(s => s.n));
const unknown = cited.filter(n => !valid.has(n));

if (declined) grounded = false;
else if (cited.length === 0) grounded = false;
else if (unknown.length) grounded = false;

What you import

One workflow file with 28 nodes. Setup notes are pinned beside every step, and your Drive folder is re-indexed whenever a document changes.

The complete workflow on the n8n canvas, with its sticky-note setup guide
n8n, self-hosted or Cloud A Google Drive folder OpenAI or any compatible API Telegram, or swap in Slack

Running in about 15 minutes

  1. ImportChoose Import from file in n8n.
  2. Pick your folderSelect your docs folder in the two Drive triggers.
  3. ConnectAdd Drive, OpenAI and Telegram credentials. Both embedding nodes use the same model.
  4. TuneStart with the default MIN_SCORE, then adjust on real questions.
  5. TestAsk one question your docs answer and one they don't.

Questions

What exactly do I get?

One n8n workflow file (JSON, 28 nodes) with setup notes pinned inside it. No account, subscription or hosted service. It runs in your own n8n.

Can I use it for client work?

Yes, in as many of your own or your clients' workflows as you like. You can't resell or republish the template itself.

Does the price include AI costs?

No. It uses your own OpenAI or compatible key, so usage is billed by that provider. Embedding a typical FAQ costs cents.

Can I make it stricter?

Raise MIN_SCORE to escalate more, or require a citation per paragraph in the citation check. For production, swap the built-in vector store for Qdrant, Supabase or PGVector.

What if it doesn't work for me?

Email hello@hundredships.com for help, or ask for a full refund within 14 days.