Turn JSON files on your FTP server into a searchable knowledge index. The flow reads files, creates text embeddings with OpenAI, and stores them in Qdrant for fast semantic search. It suits teams building chatbots, help centers, or internal search tools.

The run starts with a manual test trigger in n8n. It lists all JSON files on FTP, loops through each file, and downloads the binary content. A data loader converts the JSON into a document format. If needed, a character text splitter breaks the content into smaller chunks using a chunk id marker. OpenAI then turns each chunk into a vector. The vectors and metadata go to a Qdrant collection in batches of 100 using cosine distance and 1536 vector size.

You need FTP access, an OpenAI API key, and a Qdrant endpoint with a collection that matches the 1536 vector size. Expect a consistent index that updates from your file source with one click. Common uses include product docs, policy libraries, and multilingual content. Most teams cut manual prep time and gain better search results without writing custom code.

• FTP listing of all JSON files with a simple path setting
• Loop through files one by one using Split In Batches for stable processing
• Binary file download for each item to ensure full data capture
• Default Data Loader converts JSON into documents ready for embedding
• Character Text Splitter optionally splits content by chunk id for cleaner chunks
• OpenAI Embeddings create 1536 dimension vectors for each chunk
• Qdrant Vector Store inserts vectors and metadata into a chosen collection
• Batch size of 100 improves throughput while keeping resource use steady

• Reduce manual file handling from hours to minutes by automating listing, download, and upload steps
• Streamline content ingestion by up to 80 percent with batch processing and looping
• Improve search relevance by using semantic vectors instead of basic keywords
• Handle 10 times more documents by processing 100 items per batch
• Connect legacy FTP storage with a modern vector database without custom code

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 FTP, OpenAI and Qdrant. See the Tools Required section above for links to create accounts with these services.
3. In the n8n credentials manager, create FTP credentials with your host, port, username, and password. Test the connection to confirm access to your JSON directory.
4. Open the FTP List node and set the folder path that contains your JSON files. Save the node and run a test to see the file names.
5. Open the Split In Batches node and keep the default of one item per loop to process files safely.
6. Open the FTP Download node and confirm the path uses the file name from the loop. Make sure the binary property is set to binary.data.
7. Open the Default Data Loader node and set data type to binary so the JSON file is converted into a document format for embedding.
8. If your files include chunk markers, open the Character Text Splitter and keep the separator as chunk_id. If not, adjust the separator or skip splitting.
9. In credentials, create an OpenAI API Key by pasting the key from your OpenAI account. In the Embeddings node, pick the embedding model that uses 1536 dimensions.
10. Create Qdrant credentials with your API key and endpoint. In the Qdrant node, select your collection and keep batch size at 100. Ensure the collection uses size 1536 and cosine distance.
11. Click Execute Workflow to run a manual test. Check node outputs for errors and confirm vectors appear in your Qdrant collection.
12. If you see missing files or empty outputs, verify the FTP path and file extensions. If Qdrant rejects inserts, confirm vector size matches the embedding model.