Get faster replies out of your support inbox without losing control. The flow reads new messages, summarizes them, pulls the right facts from your knowledge base, writes a short answer, and asks for a simple approval before sending. It suits teams that handle many inbound questions and want consistent, on brand replies.

Here is how it works. New emails arrive through IMAP. The body is converted to Markdown so the models can read it clearly, then a summarization chain powered by DeepSeek creates a short brief. A Qdrant vector store, filled with your company documents using OpenAI embeddings, is queried to fetch helpful context. OpenAI writes a reply under 100 words. The draft is sent to Gmail where you approve with a YES or NO. If approved, the message goes out via SMTP with the original subject and recipient.

To set it up, you need Gmail, IMAP and SMTP access, OpenAI, OpenRouter for the DeepSeek model, and a Qdrant collection with your documents. Add the collection name and API keys, then test with a real email. Teams typically cut handling time per message and keep answers consistent across agents. Good fits include product questions, policy requests, and simple sales inquiries that need fast, accurate replies.

• IMAP intake reads new emails from your support inbox
• HTML to Markdown conversion so models can parse the message cleanly
• Email summarization chain using DeepSeek to create a short brief of the message
• Qdrant vector search retrieves company facts using OpenAI embeddings
• OpenAI composes a professional reply capped at 100 words
• Gmail draft is created and waits for a YES or NO approval
• Approval gate sends the final email via SMTP to the original sender
• Subject and recipients are preserved using Re and the original from and to fields

• Reduce manual handling time per email from 10 minutes to about 2 minutes with draft and approve flow
• Automate up to 70 percent of routine replies while keeping human approval on final send
• Lower response times by 60 percent through instant summaries and suggested replies
• Improve answer consistency by sourcing facts from one company knowledge base
• Handle up to 5 times more inbound emails without adding headcount
• Connect email intake and knowledge retrieval in one place for simpler operations

1. Import the template into n8n: Create a new workflow in n8n > Click the three dots menu > Select 'Import from File' > Choose the downloaded JSON file.
2. You'll need accounts with Gmail, OpenAI, Qdrant, OpenRouter and IMAP. See the Tools Required section above for links to create accounts with these services.
3. In n8n, open the Email Trigger IMAP node. In the credential field, click Create new credential and enter your IMAP server, port, username, password, and SSL settings. Save and test by pulling one recent email.
4. Open the Gmail Send Draft node. Click Create new credential and complete the Gmail OAuth2 flow. Choose the mailbox that will hold drafts and support send and wait for response.
5. Open the Send Email SMTP node. Create a new SMTP credential with your server, port, username, password, and secure setting. Send a test email to verify delivery.
6. Open the OpenAI nodes. For both Embeddings OpenAI and OpenAI chat, click Create new credential and paste your OpenAI API key from the OpenAI API page. Name the credential clearly so you can reuse it.
7. Open the DeepSeek R1 node. Click Create new credential for OpenRouter and paste your OpenRouter API key from your OpenRouter account. Select the deepseek r1 model in the node.
8. Configure the Qdrant Vector Store node. Create a new Qdrant API credential with your Qdrant URL and API key. Set the collection name to match your existing collection and keep mode as retrieve as tool.
9. If you have not loaded documents into Qdrant, prepare your content and embed it once using your preferred method so the collection is ready. Make sure the collection name matches the node field.
10. Check the Markdown node mapping. The html field should read the incoming email body. Keep the default to make sure the model sees clean text.
11. Review the Approve decision node. Confirm the condition listens for a YES response from the Gmail draft thread and routes approved messages to Send Email.
12. Run a full test: send a sample email to the IMAP inbox, wait for the Gmail draft, reply YES to approve, and confirm the final email is delivered to the original sender with the Re subject.
13. Troubleshoot common issues: if IMAP does not pull, verify server and SSL; if the draft does not appear, recheck Gmail OAuth and selected account; if Qdrant returns no results, confirm the collection name and document embeddings; if OpenAI errors, verify API keys and model limits.