Review: Vector Store Tool (+ Langchain Code Alternative!)

Hello, is it possible that the search filter on the Qdrant vector store (retrieve) won't work when connected to "Vector Store Without Chain"?

Yes, you will need to update to release 1.50.0 or later, as the search filter option was introduced in the Qdrant vector store node.

image637×647 19.3 KB

Yes, I have it, but the filter isn't functioning when connected to that Langchain code.

Interesting… I tested again with all 3 examples in this post and it does seem to be working for me at least. Perhaps you’ve uncovered a bug?

If you are able to share your workflow and maybe some sample data (e.g., how are you defining metadata?), that would be very helpful!

Hi, I've just added metadata to a collection in Qdrant and am using a filter like this:

json
{
"should": [
{
"key": "metadata.CaseID",
"match": {
"text": "SOMETHING1"
}
},
{
"key": "metadata.CaseID",
"match": {
"text": "SOMETHING2"
}
}
]
}


When the Qdrant Vector Store node is connected to the Vector Store Retriever, the filter works correctly. However, if it's connected to the Vector Store Without Chain, it does not.

I have the collection set up with this simple information:

image547×626 15.2 KB

Thanks for the example. This is quite similar to what I used for testing, so I'm not entirely sure why our outcomes differ.

Based on what you shared, I suspect that match only accepts value, any, and except as child keys (see docs), which might make text invalid.

Could you please try one of these options:

{
  "should": [
    {"key": "metadata.CaseID", "match": { "value": "SOMETHING1" } },
    {"key": "metadata.CaseID", "match": { "value": "SOMETHING2"} }
  ]
}

or using the any option:

{
  "must": [
      {
          "key": "metadata.CaseID",
          "match": {
              "any": [ "SOMETHING1", "SOMETHING2" ]
          }
      }
    ]
}

My second build functions flawlessly for my specific needs. However, I'm surprised this isn't included as a node option. Thank you!

Sorry but hijacking this lol… Ok… the whole stuff works… but when I get to the vector store (qdrant) I want to use the filter with the project_id that the agent got from another tool.

I’m trying to figure out for myself for 2 days how can I pass something to that filter field but the only thing it sees is the chat node can't see anything the agent node produced… How would you approach that? my workflow is basically like yours chat, agent, one tool to retrieve the project one to go to the vector store that uses a qdrant vector store with open ai embedding and model

Found in another topic. I've added an additional agent before the main agent to recover the ID. This involves one more call to the AI API, but it allows me to reference it in the vector store filter.

Hello,

This is a known limitation of the Vector Store Tool; you cannot modify this value directly within the agent node. However, it is indeed possible for the agent to utilize values from one tool for its other tools.

Here is a concise template to illustrate the usage with the Langchain Code node:

• In the chat, prompt with: "I’m looking for a spicy Asian recipe."
• The agent will employ the cookbook tool to identify available cookbooks and retrieve their corresponding IDs.
• Subsequently, the agent will utilize the cookbook recipe tool, passing the most suitable cookbook ID along with the user's query.

Does this mean we can connect to a vector store not currently listed? Many databases now offer their own vector capabilities. I'm trying to understand how I might use that instead of what callin.io provides. Thanks

Does this imply we can integrate with a vector store not currently listed?

To utilize a vector store not supported by Langchain, yes, it's possible by employing Langchain's DynamicStructuredTool. This requires:

1. You are using a self-hosted version of callin.io.
2. You have a method to connect to your vector store using JavaScript (e.g., REST, RPC, GraphQL).

Additionally, if your vector store is supported by Langchain but not yet available in callin.io, consider submitting a Feature Request to the development team.

Thanks Jim. Yes, the vector store is Couchbase (if you're familiar with it). It's supported by Langchain, but not directly exposed by callin.io. A request has been made. Do you have an estimate on when this might be addressed?

Will the workaround involve using DynamicStructuredTool exclusively?

Do you know approximately when this will be addressed?

Given the volume of other requests, I don't expect it very soon, but it's good to note that people are asking for this functionality!

Will the workaround involve solely using DynamicStructuredTool?

Yes. I've completed a comparable task for Redis. You can find the tutorial here -Using Redis VectorStore Search in callin.io