Give customers fast travel advice with a chat assistant that uses your own points of interest. It stores your data in Couchbase and answers questions with AI, so teams can guide travelers with less manual work.

A chat message starts the assistant. It uses a Google Gemini chat model with short memory to hold context, and it calls a vector search tool to find the best matches from Couchbase. New places are added by a webhook that ingests text, splits it if needed, creates OpenAI embeddings, and inserts vector data into Couchbase. During chat, the agent retrieves the top 10 results by similarity and uses them to answer. The system message tells the agent to trust the database and not make things up.

Setup needs a Google API key for Gemini, an OpenAI API key, and a Couchbase cluster with a vector search index. Allow your n8n IP, create the bucket, scope, and collection, and import the index JSON. Send the sample curl request to load places, then ask questions like romantic getaways or family friendly spots. Expect faster answers, less research time, and better use of your own content for travel teams and tourism groups.

• Chat trigger starts the assistant when a message arrives.
• Google Gemini chat model generates clear and helpful replies.
• Memory window keeps recent chat context for better answers.
• Agent tool calls Couchbase vector search to fetch relevant places.
• OpenAI embeddings turn text into vectors for search and retrieval.
• Webhook ingests new points of interest with a simple JSON body.
• Text splitter prepares long text for better embeddings and search.
• Couchbase insert and search nodes manage vector data at scale.

• Reduce manual research from 60 minutes to 3 minutes per request
• Automate up to 80 percent of destination questions with your own data
• Improve answer accuracy by 40 percent by grounding replies in Couchbase
• Handle 50 plus concurrent chats without extra staff
• Connect AI search and data ingestion in one flow
• Speed up content updates by posting new places through a simple webhook

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 Google Gemini, OpenAI, Couchbase, Webhook.site and Postman. See the Tools Required section above for links to create accounts with these services.
3. In the n8n credentials manager, create a Google Gemini credential. Use the API key from your Google AI account. Then open the Google Gemini Chat Model node and select this credential.
4. Create an OpenAI API key in your OpenAI account. In n8n, add an OpenAI credential and select it in both OpenAI Embeddings nodes. Keep the model as text-embedding-3-small.
5. Set up Couchbase: create a cluster, add a bucket travel-agent, a scope vectors, and a collection points-of-interest. Add a database user with read and write access. Allow your n8n IP in the Couchbase console.
6. Import the provided search index JSON in the Couchbase Search section. Note the index name and confirm it matches in both Couchbase vector nodes in n8n.
7. Open the Couchbase Insert and Retrieve nodes in n8n. Create a new Couchbase credential if prompted, select your bucket, scope, collection, and the vector index name. Save the node settings.
8. Check the Webhook node. Confirm it is POST and Raw Body is enabled. Copy the Production Webhook URL from the node header after you activate the workflow.
9. Activate the workflow. Use Postman or Webhook.site to post the sample JSON body shown in the notes. Verify the Insert node runs and data appears in your Couchbase collection.
10. Open the chat interface connected to the chat trigger. Ask a question like Where should I go for a romantic getaway and review the answer.
11. If no results show, confirm the index name, the text field key description, and that embeddings are created. Ensure the OpenAI key is valid and the model name is correct.
12. If Couchbase rejects the connection, verify allowed IPs, user permissions, and the connection string in the Couchbase credential.
13. Adjust topK in the retrieve node if answers feel too broad or too narrow. Fine tune the system message in the agent to reduce made up content.