Email replies drawn from a Qdrant knowledge base

Email replies drawn from a Qdrant knowledge base

An n8n workflow that answers company questions sent by email, from your own documents. One part, which you start by hand,
reads a Google Drive folder and stores the documents in a Qdrant collection. The other part watches a mailbox over IMAP,
summarizes each new email, writes a reply with an AI agent that searches the collection, and sends it by SMTP.

What is inside

• workflow.json: the workflow as n8n exports it, 26 nodes. Three are sticky notes that label the steps. It holds no
  credential.
• README.md: these notes.

What we changed from the original copy

• The two DeepSeek model nodes named deepseek/deepseek-r1:free, which OpenRouter no longer lists. They now name
  deepseek/deepseek-r1-0528.
• Create collection sent a POST with an empty filter, which does not create a collection. It now sends a PUT with a vector
  size of 1536 and the Cosine distance, and it is set to continue if the collection already exists.
• The folder ID in Get folder was a test folder of the author's. It is cleared, so choose yours.
• Every credential link is gone, so select your own in each node that needs one.
• Nothing else. The QDRANTURL and COLLECTION placeholders are as the original had them.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Qdrant instance, on Qdrant Cloud or your own server, with its address and an API key.
• An OpenAI API key, and an OpenRouter API key for the DeepSeek nodes (or an OpenAI model in their place).
• A Google Drive folder of Google Docs that answer the questions you get.
• A mailbox with IMAP and SMTP details, ideally a test one to start with.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. In Create collection and Refresh collection, replace QDRANTURL with your Qdrant host (with its port, if it needs one)
   and COLLECTION with a collection name of your choice. Add a Header Auth credential with the name api-key and your Qdrant
   key as the value, and select it in both nodes.
3. Add a Qdrant API credential and select it in Qdrant Vector Store and Qdrant Vector Store1. In both, replace
   COLLECTION with the same collection name.
4. Add an OpenAI credential and select it in OpenAI, OpenAI 4-o-mini, Embeddings OpenAI and Embeddings OpenAI1.
5. Add a second OpenAI credential whose base URL is https://openrouter.ai/api/v1, with your OpenRouter key, and select it in
   DeepSeek R1 and DeepSeek. Or pick an OpenAI model in both and use the first credential.
6. Add a Google Drive OAuth2 credential and select it in Get folder and Download Files. In Get folder, choose your
   folder.
7. Run the first part once from When clicking ‘Test workflow’. It creates the collection, clears it, then reads the folder,
   splits each document and stores the pieces.
8. Add an IMAP credential and select it in Email Trigger (IMAP). Add an SMTP credential and select it in Send Email.
9. Send a test question from another address, one that your documents answer. Read the reply, then press Publish.

How a run goes

The first part, started by hand:

1. When clicking ‘Test workflow’ starts Create collection and then Refresh collection, which sends a delete-points
   request with an empty filter before each fill.
2. Get folder lists the files in your folder, and Download Files downloads each one as plain text.
3. Default Data Loader and Token Splitter cut each file into pieces of 300 tokens with an overlap of 30.
4. Embeddings OpenAI1 turns the pieces into vectors, and Qdrant Vector Store1 stores them.

The second part, for each new email:

1. Email Trigger (IMAP) picks up an unread email from the INBOX and marks it read.
2. Markdown converts the email's HTML, and Email Summarization Chain writes a summary of up to 100 words.
3. Email Classifier keeps summaries of emails that ask about the company. The rest go to Do nothing.
4. Write email is an agent. It searches the collection through Qdrant Vector Store and writes a reply from the summary.
5. Review email formats the reply as HTML.
6. Send Email sends it to the sender, from the address that received the email, with the subject Re: and the original
   subject.

Change the result

• Which emails get an answer: the categories in Email Classifier.
• How the reply reads: the system message of Write email.
• The look and length of the reply: the prompt in Review email.
• Which documents it reads: the folder in Get folder.
• How documents are cut: Token Splitter.
• The models: OpenAI, OpenAI 4-o-mini, DeepSeek R1 and DeepSeek. Change the embedding model in both embedding nodes
  and the vector size in Create collection together.

Good to know

• OpenAI and OpenAI 4-o-mini call gpt-4o-mini. OpenAI's deprecations page, read on 12 November 2026, lists it for
  shutdown on 14 January 2027, with gpt-6-luna as its replacement.
• deepseek/deepseek-r1-0528 is a paid model on OpenRouter (its model list, read the same day), unlike the free one the file
  named.
• The stored pieces and the searches must use the same embedding model. Both embedding nodes use the node's default,
  text-embedding-3-small, which has 1536 dimensions, the size Create collection now asks for.
• Create collection is set to continue on error, so a collection that exists already does not stop a rerun. A wrong address
  or key shows up in Refresh collection instead.
• Refresh collection is meant to clear the collection before each fill. Try it on a throwaway collection first.
• The reply is sent as soon as it is written, from the address that got the email, with no review. Test with your own address.
• Nothing checks who sent an email, so an automated sender that asks a company question gets a reply too.
• Write email reads the summary, not the email, so details the summary drops do not reach the reply.
• Markdown converts the email's HTML part. Test with the kinds of email you receive, plain text as well.
• Get folder returns up to 50 files, and Download Files converts Google Docs to plain text.
• Email text goes to OpenAI and OpenRouter, and your documents go to OpenAI to be turned into vectors.