Supabase metadata

[Dilbert]

I’m trying to pass metadata to the supabase vector store node to save with the embeddings. this includes location data and client number.

The table is set up with a location column and it is ready for geographical information. Each time i pass the data the meta data is added as a row in the table and not to the column location, does anyone have any pointers to how I can correctly set this up?

[amirfahd]

Do you have RLS enabled? Maybe this is a policy issue. If you’re seeing location data getting passed but not being stored correctly you can check on your RLS policy.

Another thing could be a mismatch of data type.

[Dilbert]

No RLS is disabled until i can figure out how to get the metadata written, the mismatch of data type is a possibility for the locaiton although the client_id i’m also trying to pass is just a text field.

[amirfahd]

Okay, there could be an issue in parsing

Possible json format being passed

{
  "metadata": {
    "location": "POINT(125.607 7.072)",
    "client_id": "abc123"
  }
}

You need to flatten your metadata

{
  "location": {
    "type": "Point",
    "coordinates": [125.607, 7.072]
  },
  "client_id": "abc123"
}

Or if you are using SQL or a prepared payload in n8n

{
  "location": "POINT(125.607 7.072)", // as string, if your column is text
  "client_id": "abc123"
}

[amirfahd]

You can check this, this might help you @Dilbert

Multiple entries in supabase vector store metadata filter

Hello everyone, I’m currently working on a RAG cha where users can access a set of meetings, and I need to filter the available meetings based on the meeting_ids that the user has access to. Here’s the scenario: i have a vector store with the my metadata looking like this: { meeting_id: "id1", other_fields: "blah"} Each user has a list of meeting_ids that define which meetings they have access to. These meeting_ids are hard-coded as of now, just to test. When filtering the meetings that a u…

[Dilbert]

@amirfahd

  "location": {
    "type": "Point",
    "coordinates": [125.607, 7.072]
  },
  "client_id": "abc123"
}

Thanks for your help amirfahd. The metadata is going into the database just not in the right columns

The code I’m using before the supabase vector store node is return
[
{
json: {
embedding: [0.12, 0.88, 0.45],
consultant_id: “abc-123”,
location: {
type: “Point”,
coordinates: [-0.1278, 51.5074]
},
}
}
];

Any Ideas?

[amirfahd]

Okay this is good. You just need to add a Code node before saving to the Supabase vector to flatten the json being passed.

The thing is your input:

{
  json: {
    embedding: [0.12, 0.88, 0.45],
    consultant_id: “abc-123”,
    location: {
      type: “Point”,
      coordinates: [-0.1278, 51.5074]
    },
  }
}

We should flatten it to exactly match the Supabase columns which it doesn’t at the moment.

Add this code inside the Code Node and let me know how it goes.

return $input.all().map(item => {
  const { text, embedding, consultant_id, location, metadata } = item.json;

  const [lon, lat] = location.coordinates;

  return {
    json: {
      content: text,
      embedding: embedding,
      consultant_id: consultant_id,
      location: `POINT(${lon} ${lat})`,
      metadata: metadata || {}
    }
  };
});

Your json should look like this after passing the Code node.

{
  "content": "This is a sample document",
  "embedding": [0.12, 0.88, 0.45],
  "consultant_id": "abc-123",
  "location": "POINT(-0.1278 51.5074)",
  "metadata": {
    "source": "scraped",
    "priority": "high"
  }
}

Have a great day!

[Dilbert]

@amirfahd

{
  "content": "This is a sample document",
  "embedding": [0.12, 0.88, 0.45],
  "consultant_id": "abc-123",
  "location": "POINT(-0.1278 51.5074)",
  "metadata": {
    "source": "scraped",
    "priority": "high"
  }
}

Amirfahd,

I think it is the supabase vector store node that is causing me the issue, if I use the supabase node i get the correct data written to the columns, if I use the vector store node I do not, it just goes as another row.

I need to be able to store the id and the location on each row of the embeddings, any further ideas?

Thanks

Andy

[amirfahd]

Hi @Dilbert,

What you can do is just do as is,

1. Use the Supabase Vector Store to insert embedding, content, and maybe metadata
2. Then add a Supabase Update node using the returned id
3. Set location and consultant_id correctly in that row

Let me know if that solved your issue.

[King_Samuel_David]

nice ideas, need to use superbase more thanks for this info