Support for ElasticSearch as a vector store

[vinodkiran]

No description provided.

[HenryHengZJ]

Were you able to get it working? @vinodkiran

[vinodkiran]

@HenryHengZJ

There seems to be an issue with the underlying vector store implementation. I have raised a ticket with the langchain team.

langchain-ai/langchainjs#2857

[vinodkiran]

@HenryHengZJ PR Is now complete...Kindly review and merge.

[HenryHengZJ]

I have the following error:

[vinodkiran]

I have the following error:

Are you using the Retrieve Node? Can you add an image of the entire flow?

[vinodkiran]

The Underlying langchain vector store creates the following index. Kindly check your index in Elastic Cloud.

{
  "mappings": {
    "dynamic_templates": [
      {
        "metadata.*": {
          "match_mapping_type": "*",
          "mapping": {
            "type": "keyword"
          }
        }
      }
    ],
    "properties": {
      "embedding": {
        "type": "dense_vector",
        "dims": 1536,
        "index": true,
        "similarity": "l2_norm",
        "index_options": {
          "type": "hnsw",
          "m": 16,
          "ef_construction": 100
        }
      },
      "metadata": {
        "properties": {
          "source": {
            "type": "keyword"
          }
        }
      },
      "text": {
        "type": "text"
      }
    }
  }
}

[HenryHengZJ]

ah it works now:

Previously I created the index myself on ElasticSearch and that ran into error. So I specify a new index name that is not being created yet and have the underlying langchain implementation to create that index, that works.

[HenryHengZJ]

much appreciated for figuring out the issue and got it working!

[HenryHengZJ]

@vinodkiran the metadata filter is still not working right?

we can open another PR to enhance that