How to add metadata to all chunks when embedding in Pinecone Vector Database

[koala_ty]

Hi,

I am trying to embed text chunks with metadata (document_type and date) into a Pinecone Vector Database. I have tried using the PInecone Vector Store but with no success. How can I include the metadata in all the chunks that I embed into the database?

[jennapederson]

Can you share more about your setup, which data loader/text splitter you’re using? Are you experiencing any errors or are the vectors being upserted but no metadata is shown in the Pinecone console?

You should be able to set metadata right on the data loader node like this:

I just tested this out on my end and it’s working like this, but let me know your setup and I can try to help.

[koala_ty]

Hi Jenna,

Thanks for the response. I’m using Default Data Loader and Recursive Text Splitter. Let me try out what you did and ask further questions if I have any.

[koala_ty]

@jennapederson I think n8n has a limitation where a user can only use ‘eq’ as an operator to filter on PInecone DB Metadata, correct? See below forum link. I wanted to use the time/date as metadata and filter on “everything since October 5” as an example. Can I do that through n8n and Pinecone?

Pinecone Vector Store in N8n only supports $eq in metadata filters — why?

Pinecone Vector Store in N8n only supports $eq in metadata filters — why? Hi team I’m currently using the Pinecone Vector Store: Retrieve Documents (As Tool For AI Agent) node in a workflow to retrieve vectorized documents filtered by metadata. It works fine when using simple filters like: json CopiarEditar { "group_id": { "$eq": 3 } } However, I’ve hit a significant limitation: the node does not seem to support other valid Pinecone operators such as $in,…

[jennapederson]

I remember that being an issue. It looks like the node just assumes key/value pair rather than passing along an operator. On the backend, we support both `{"genre": "documentary"}` and`{"genre": {"$eq": "documentary"}}`and I suspect the node is using the first one by default.

I absolutely understand using the existing node would be better, but you could try implementing the query using an HTTP node and using this curl. You’d also have to generate your own query embedding first to pass into that vector field (that’s something the existing node does for you).

# To get the unique host for an index,
# see https://docs.pinecone.io/guides/manage-data/target-an-index
PINECONE_API_KEY="YOUR_API_KEY"
INDEX_HOST="INDEX_HOST"

curl "https://$INDEX_HOST/query" \
  -H "Api-Key: $PINECONE_API_KEY" \
  -H 'Content-Type: application/json' \
  -H "X-Pinecone-API-Version: 2025-04" \
  -d '{
        "vector": [0.0236663818359375,-0.032989501953125,...,-0.01041412353515625,0.0086669921875],
        "namespace": "example-namespace",
        "topK": 3,
        "filter": {"genre": {"$eq": "documentary"}},
        "includeMetadata": true,
        "includeValues": false
    }'

You can read more about this here: Filter by metadata - Pinecone Docs

Thanks for raising this. I’m adding this to a list of items we’re tracking for improvements with the Pinecone Vector Store node.

[koala_ty]

Hi Jenna, thanks for your help! Couldn’t I also use a Javascript Code Node and do the query embedding + Pinecone query in the same, one Code Node?

[jennapederson]

I don’t think you’ll have access to external dependencies like the Pinecone typescript SDK. So I think you could use a Code node that makes an HTTP call (similar to an HTTP node), but you’re right, you’d have to add in the query embedding too.

Just remember that you’d have to store a timestamp value as a number (unix timestamp) as metadata as that’s what the $ge operates on.

[koala_ty]

None of the Code commands ($http, fetch, axios) are working to make the API calls, so I’ll probably have to use the n8n HTTP node.

I’m storing the timestamp as an integer number in the format YYYYMMDD since I only care about the day my data happened on.